Provides AI clients with over 300 local developer utilities including JSON processing, encoding, hashing, regex, and token management.
Also known as utilix-tech/utilix-mcp, utilix-mcp, @utilix-tech/mcp
- Stars
- 0
- Tools
- 300
- Downloads, 30 days
- —
- Latest version
- 0.109.0
- Last commit
- Nov 16, 2026
- License
- MIT
What is Utilix MCP?
Utilix provides AI clients with over 300 local developer utilities including JSON processing, encoding, hashing, regex, and token management. It exposes 300 tools, such as format_json, minify_json, diff_json.
Summary generated from the project's README
About Utilix MCP server
Utilix provides a collection of over 300 deterministic developer utilities designed to work with MCP-compatible AI clients like Claude Desktop, Cursor, VS Code Copilot, and Windsurf. These tools run locally on your machine via npx, meaning they do not require any API keys or external accounts to function. This setup ensures that sensitive operations like PII detection or secret scanning happen within your local environment.
The server includes specialized toolsets for various programming tasks, such as JSON processing, encoding, hashing, and regex testing. It features a dedicated suite of 28 AI agent utilities specifically for LLM pipelines, including token estimation, text chunking, and context compression. Developers can use these tools to manage token budgets, extract structured data from messy text, or validate schemas without needing to write custom scripts for every small utility task.
What you can do with it
- 1Manage LLM token budgetsAsk the AI to trim a long text to a specific token count to ensure it fits within a model's context window.
- 2Secure sensitive dataInstruct the client to identify and redact personally identifiable information from a dataset before logging it.
- 3Perform structural data comparisonsRequest a comparison between two JSON configurations to identify specific changes in paths or values.
- 4Optimize web content for ingestionHave the AI strip unnecessary HTML elements to reduce the number of tokens used when processing a webpage.
Written from the project's README. Every use case is checked against a quote from it.
How to install Utilix MCP
Formats verified 10 Nov 2026Add to CursorAdd to GooseInstall Utilix MCP in Claude Code
$ claude mcp add --env 'UTILIX_API_KEY=<YOUR_UTILIX_API_KEY>' utilix-mcp -- npx -y @utilix-tech/mcpInstall Utilix MCP in Cursor
~/.cursor/mcp.json
{
"mcpServers": {
"utilix-mcp": {
"command": "npx",
"args": [
"-y",
"@utilix-tech/mcp"
],
"env": {
"UTILIX_API_KEY": "<YOUR_UTILIX_API_KEY>"
}
}
}
}Install Utilix MCP in VS Code
.vscode/mcp.json
{
"servers": {
"utilix-mcp": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"@utilix-tech/mcp"
],
"env": {
"UTILIX_API_KEY": "<YOUR_UTILIX_API_KEY>"
}
}
}
}Install Utilix MCP in VS Code CLI
$ code --add-mcp '{"name":"utilix-mcp","type":"stdio","command":"npx","args":["-y","@utilix-tech/mcp"],"env":{"UTILIX_API_KEY":"<YOUR_UTILIX_API_KEY>"}}'Install Utilix MCP in Claude Desktop
claude_desktop_config.json
{
"mcpServers": {
"utilix-mcp": {
"command": "npx",
"args": [
"-y",
"@utilix-tech/mcp"
],
"env": {
"UTILIX_API_KEY": "<YOUR_UTILIX_API_KEY>"
}
}
}
}Install Utilix MCP in Codex
$ codex mcp add utilix-mcp --env 'UTILIX_API_KEY=<YOUR_UTILIX_API_KEY>' -- npx -y @utilix-tech/mcpInstall Utilix MCP in Codex config.toml
~/.codex/config.toml
[mcp_servers.utilix-mcp] command = "npx" args = ["-y", "@utilix-tech/mcp"] [mcp_servers.utilix-mcp.env] UTILIX_API_KEY = "<YOUR_UTILIX_API_KEY>"
Install Utilix MCP in Gemini CLI
$ gemini mcp add -s user -e 'UTILIX_API_KEY=<YOUR_UTILIX_API_KEY>' utilix-mcp npx -- -y @utilix-tech/mcpInstall Utilix MCP in Gemini settings.json
~/.gemini/settings.json
{
"mcpServers": {
"utilix-mcp": {
"command": "npx",
"args": [
"-y",
"@utilix-tech/mcp"
],
"env": {
"UTILIX_API_KEY": "<YOUR_UTILIX_API_KEY>"
}
}
}
}Started by HostedMCP on 16 Nov 2026
We ran this command for @utilix-tech/mcp 0.109.0 in an isolated container with no network and no keys. It answered and listed 300 tools.
Utilix MCP tools 300
Read from the server with tools/listformat_jsonParse, validate and pretty-print a JSON string (strict RFC 8259). Indent with spaces or tabs, or minify with indent 0; optionally sort keys alphabetically at every level. Invalid JSON returns a specific error (trailing comma, unquoted key, single quotes, missing comma, comment, unclosed bracket) with its line and column. Valid JSON also lists warnings for duplicate keys and integers too large for a JavaScript number.
minify_jsonMinify a JSON string by removing all whitespace.
diff_jsonCompare two JSON strings and return a structured diff showing added, removed, and changed keys.
json_to_csvConvert a JSON array of objects to CSV format.
csv_to_jsonConvert CSV text to a JSON array of objects.
jsonpathEvaluate a JSONPath expression against a JSON string and return matching values.
validate_jsonValidate a JSON string against a JSON Schema (Draft-7). Returns pass/fail with error details.
yaml_to_jsonConvert a YAML string to JSON.
json_to_yamlConvert a JSON string to YAML.
json_to_typescriptGenerate TypeScript interfaces from a JSON object.
json_to_goGenerate Go struct definitions from a JSON object.
json_to_pythonGenerate Python TypedDict, dataclass, or Pydantic models from a JSON object.
Show all 300 toolsShow fewer tools
json_to_zodGenerate a Zod schema from a JSON object.
base64_encodeEncode a string to Base64.
base64_decodeDecode a Base64 string.
url_encodePercent-encode a URL or URL component.
url_decodeDecode a percent-encoded URL or URL component.
html_entities_encodeEncode special characters as HTML entities (e.g. < → <).
html_entities_decodeDecode HTML entities back to plain text (e.g. < → <).
base32_encodeEncode a string to Base32.
base32_decodeDecode a Base32 string.
hashHash a string using MD5, SHA-1, SHA-256, SHA-384, or SHA-512.
hash_allHash a string with all algorithms (MD5, SHA-1, SHA-256, SHA-384, SHA-512) at once.
bcrypt_hashHash a password using bcrypt.
bcrypt_verifyVerify a password against a bcrypt hash.
generate_uuidGenerate one or more UUIDs (v4 random or v7 time-ordered).
generate_ulidGenerate a ULID (Universally Unique Lexicographically Sortable Identifier).
generate_passwordGenerate a cryptographically secure random password.
convert_caseConvert a string to a different case format: camelCase, snake_case, kebab-case, PascalCase, SCREAMING_SNAKE, Title Case, etc.
slugifyConvert text to a URL-friendly slug.
word_countCount words, characters, sentences, and estimate reading time for a block of text.
escape_stringEscape special characters in a string for use in JSON, JavaScript, HTML, URL, or CSV.
unescape_stringUnescape special characters in a JSON, JavaScript, HTML, URL-encoded, or CSV string.
lorem_ipsumGenerate Lorem Ipsum placeholder text.
unix_timestampConvert a Unix timestamp (seconds or milliseconds) to a human-readable date.
parse_cronParse a cron expression and return a human-readable description.
cron_next_runsShow the next N scheduled run times for a cron expression.
date_diffCalculate the difference between two dates.
convert_timezoneConvert a date/time from one timezone to another.
convert_bytesConvert a byte value between B, KB, MB, GB, TB, KiB, MiB, GiB, TiB.
px_to_unitsConvert a px value to rem, em, pt, and vw units.
decode_jwtDecode a JWT token and display header, payload, and expiry info. Does not verify signature.
build_curlBuild a cURL command from URL, method, headers, and body.
curl_to_codeConvert a cURL command to code: fetch, axios, python-requests, go-http, php-curl, or ruby-net-http.
convert_colorParse a color (HEX, RGB, HSL, HSV, or named CSS color) and return all format conversions.
format_sqlFormat a SQL query with proper indentation and keyword uppercasing.
format_htmlFormat and indent HTML markup.
test_regexTest a regular expression against an input string and return all matches with capture groups.
validate_yamlValidate YAML syntax and return stats or error details.
resolve_yaml_anchorsExpand YAML anchors, aliases, and merge keys inline, and list which anchors were referenced vs. left unused.
find_yaml_duplicate_keysParse a YAML document and report every mapping key that collides with an earlier key in the same mapping scope, with the line and column of each repeated occurrence.
toml_to_jsonConvert a TOML string to JSON.
xml_to_jsonConvert an XML document to JSON.
text_diffDiff two text blocks line-by-line and show added, removed, and unchanged lines.
html_to_markdownConvert HTML to Markdown.
line_operationsSort, deduplicate, reverse, shuffle, trim, or number lines of text.
number_to_wordsConvert a number to its English words representation, Roman numerals, or ordinal form.
unicode_inspectInspect the Unicode code point(s) of a character or string.
minify_cssMinify CSS by removing whitespace and comments.
css_gradientGenerate a CSS gradient from stops.
check_contrastCheck WCAG contrast ratio between two colors.
color_paletteGenerate a color palette (analogous, complementary, triadic, etc.) from a base color.
color_shadesGenerate a scale of shades (darker) or tints (lighter) from a base color.
cors_headersGenerate CORS response headers for a given configuration.
minify_jsMinify JavaScript by removing whitespace and comments.
format_graphqlFormat or minify a GraphQL query or schema.
parse_envParse a .env file and return all key-value pairs with validation.
parse_docker_imageParse a Docker image reference into registry, namespace, repository, tag, and digest.
generate_qrGenerate a QR code as an SVG string.
generate_random_dataGenerate random mock data: names, emails, UUIDs, phone numbers, dates, colors, IPs, etc.
parse_csvParse a CSV string and return the data as JSON.
optimize_svgOptimize and minify an SVG string.
estimate_tokensEstimate the token count and API cost for a text string across popular LLMs (GPT-4o, Claude, Gemini, etc.).
detect_piiDetect personally identifiable information (emails, phones, SSNs, credit cards, IPs, IBANs, crypto wallets) in text.
detect_secretsScan text for leaked API keys, tokens, and credentials (OpenAI, Anthropic, AWS, GitHub, Stripe, etc.).
detect_prompt_injectionDetect prompt injection and jailbreak attempts in user-provided text before passing to an LLM.
repair_jsonFix malformed JSON: unquoted keys, single quotes, trailing commas, JS comments, undefined/NaN values, unclosed brackets.
trim_to_tokensTruncate text to a token budget (end/start/middle strategy).
chunk_textSplit text into overlapping chunks for RAG.
extract_urlsExtract all URLs from text or HTML.
format_few_shot_examplesFormat a list of input/output example pairs into a standard few-shot prompt template: numbered, XML-tagged, or delimiter-separated.
sanitize_htmlStrip HTML tags, scripts, and event handlers for LLM ingestion.
flatten_jsonFlatten nested JSON to dot-notation keys.
merge_jsonDeep merge two JSON objects.
extract_jsonExtract JSON blocks from mixed LLM output text.
deduplicate_linesRemove duplicate lines from text.
extract_keywordsExtract top keywords using TF-IDF scoring.
validate_json_schemaValidate JSON data against a JSON Schema (Draft-07).
diff_json_detailedDiff two JSON objects (parsed values, not strings): returns a flat list of every path with its op (added/removed/changed/unchanged), plus counts.
extract_tablesExtract HTML tables into JSON arrays with headers and rows.
extract_entitiesExtract emails, phones, dates, IPs, currencies, hashtags from text.
compress_htmlStrip scripts, styles, classes, and whitespace from HTML to minimize tokens.
compress_markdownNormalize and compress markdown to reduce token count.
compress_jsonRemove nulls, empty arrays, and empty objects from JSON to reduce tokens.
rerank_chunksScore and rerank text chunks by TF-IDF relevance to a query.
score_relevanceScore a passage relevance to a query (0-1 score with grade).
expand_queryExpand a search query with synonyms and stemmed variants for better retrieval.
vector_similarityCompute cosine similarity, dot product, Euclidean distance, and Manhattan distance between two equal-length numeric vectors (e.g. embeddings from an external model).
calculate_retry_backoff_scheduleCompute the full sequence of delays an exponential-backoff retry policy would actually produce, given a base delay, multiplier, and max retry count, optionally capped at a max delay and with full/equal jitter ranges, plus the total best-case and worst-case wait across all retries. Distinct from build_retry_after_header, which builds a single HTTP Retry-After response header.
summarize_for_llmExtractive summarization: pick key sentences to fit a token budget.
detect_passive_voiceHeuristically flag sentences containing passive-voice constructions, e.g. "was written", "were approved by the team".
read_image_infoRead image format, dimensions, and color info directly from file bytes: no decoding required. Supports PNG, JPEG, GIF, WebP, BMP.
parse_harParse a browser DevTools .har network export into a readable list of requests (method, URL, status, size, timing) plus aggregate summary stats.
validate_sitemapValidate a sitemap.xml (urlset) or sitemap index (sitemapindex) against the sitemaps.org protocol: required <loc>, absolute URLs, valid <changefreq>/<priority>/<lastmod> values, the 50,000-URL limit, and duplicate entries.
split_sitemapSplit a URL list exceeding the sitemaps.org 50,000-URL cap into multiple sitemap.xml files plus a sitemap index file.
validate_robots_txtValidate a robots.txt file: parse it into User-agent groups and flag structural mistakes — directives before any User-agent, unknown directives, non-absolute Sitemap URLs, malformed Crawl-delay values, missing leading slashes, and duplicate rules.
generate_robots_txtGenerate a robots.txt file from one or more User-agent groups (each with Allow/Disallow rules and an optional Crawl-delay) plus a list of Sitemap URLs. The inverse of validate_robots_txt: a group with no Allow or Disallow rules gets an explicit empty Disallow line, the standard way to allow crawling everything.
validate_card_numberValidate a credit card number via the Luhn checksum and detect its network (Visa, Mastercard, American Express, Discover, Diners Club, JCB, Maestro, UnionPay) from its prefix and length. No card data is stored or transmitted.
validate_ibanValidate an IBAN via the mod-97 checksum defined by ISO 13616 and format it into the standard human-readable grouped form. Cross-checks the total length against the country's registered fixed length where known.
calculate_loanCompute the fixed monthly payment, total interest, and full month-by-month amortization schedule for a loan or mortgage from principal, annual interest rate (as a percent), and term in months.
calculate_extra_payment_payoffCompute how much faster a fixed-rate loan pays off, and how much interest is saved, when a fixed extra amount is added to every monthly payment on top of the standard amortized payment.
validate_barcodeValidate an EAN-13, EAN-8, or UPC-A barcode number via its weighted mod-10 check digit algorithm. The format is inferred from the digit length (8, 12, or 13).
calculate_taxAdd tax to a net (pre-tax) amount or remove tax from a gross (tax-inclusive) amount, given a rate. Pure formula, not a live VAT/sales-tax rate lookup.
calculate_compound_interestCompute the future value of a principal plus an optional recurring contribution under compound interest, given an annual rate (as a percent), a term in years, and a compounding frequency (periods per year), plus a year-by-year balance breakdown.
calculate_real_rate_of_returnApply the Fisher equation, (1 + nominal) = (1 + real) * (1 + inflation), to solve for whichever of the nominal rate, inflation rate, or inflation-adjusted real rate is missing, given exactly two of the three (all as percents).
validate_isbnValidate an ISBN-10 or ISBN-13 number via its checksum (inferred from the character count) and convert it to the other format where possible.
lookup_mimeLook up the MIME type for a file extension, or the file extension(s) registered for a MIME type, from a static table of well-known content types.
calculate_tip_splitCompute a tip on a bill amount and split the bill, tip, and total evenly across a number of people.
calculate_simple_interestCompute simple (non-compounding) interest as principal x rate x time, plus the total payoff. Distinct from calculate_compound_interest, which reinvests interest each period.
calculate_time_valueConvert between present and future value (time value of money). With direction "future" the amount is a present value grown to its future value at the given annual rate; with direction "present" it is a future value discounted back to present value. Compounds compoundsPerYear times per year (1 = annually).
base58_encodeEncode text as Base58 (Bitcoin-address style, no ambiguous 0/O/I/l characters).
base58_decodeDecode a Base58 string back to text.
base62_encodeEncode text as Base62 (alphanumeric, URL-safe, no punctuation).
base62_decodeDecode a Base62 string back to text.
format_ndjsonValidate newline-delimited JSON (NDJSON/JSON Lines), report per-line errors, and pretty-print each valid line.
dedupe_ndjsonRemove duplicate records from NDJSON (newline-delimited JSON) input. With no keyField, two lines are duplicates when their parsed JSON values are structurally identical. With a keyField, two lines are duplicates when that field's value matches, keeping the first occurrence of each key.
ndjson_to_csvConvert NDJSON (newline-delimited JSON, one flat object per line) into CSV, inferring the header row from the union of keys across every line.
read_pdf_metadataRead PDF metadata (title, author, dates, page count, encryption flag) directly from the trailer/Info dictionary: no PDF rendering library required.
read_pdf_page_dimensionsList each page's MediaBox width/height (points and inches), rotation, and orientation, in document order, by walking the Root -> Pages -> Kids tree: no PDF rendering library required.
read_pdf_outlineWalk a PDF's /Outlines bookmark tree (Root -> Outlines -> First/Next/First chain) and list every bookmark's title, nesting level, and link target, in document order: no PDF rendering library required. Does not resolve a page-ref target to a page number.
read_exif_dataRead common EXIF tags (camera make/model, orientation, timestamps, exposure settings, GPS coordinates) directly from a JPEG's APP1 segment. Only JPEG is supported.
parse_jwksParse a JSON Web Key Set (or a single bare JWK) and summarize each key's type, use, algorithm, key ID, and approximate size. Does not verify or use the keys cryptographically.
compute_jwk_thumbprintCompute the RFC 7638 thumbprint of a single JWK: canonicalize its required members into lexicographically-ordered, whitespace-free JSON and hash it. A fingerprint for identifying/matching a key, not a signature or verification. Accepts a bare JWK only, not a JWKS set.
read_wav_infoRead sample rate, channel count, bit depth, and duration directly from a WAV file's RIFF/fmt/data chunk headers. No audio library required.
score_readabilityScore text with Flesch Reading Ease, Flesch-Kincaid Grade Level, and Gunning Fog Index, using a heuristic syllable counter.
analyze_sentence_lengthsSplit text into sentences and report a histogram of sentence lengths in words, plus mean, median, standard deviation, and a consistency label. Distinct from readability grade scores.
read_id3_tagsRead ID3 tags (title, artist, album, year, genre, comment, track number) directly from an MP3 file's bytes. Prefers an ID3v2.3/2.4 header; falls back to a classic ID3v1/1.1 128-byte trailer. No audio library required.
read_flac_infoRead sample rate, channels, bit depth, total samples, and duration from a FLAC file's STREAMINFO metadata block, plus Vorbis comment tags (ARTIST, TITLE, ALBUM, DATE, GENRE, TRACKNUMBER) from its VORBIS_COMMENT block if present. No audio decoding library required.
read_ico_infoRead a .ico file's ICONDIR/ICONDIRENTRY header table directly from its bytes to report how many images it embeds and each one's width, height, color depth, and size. No image decoding required.
diff_env_exampleDiff a .env file against its .env.example template: flags keys missing from either side, plus keys present in both with an empty placeholder value in the example.
parse_user_agentParse a browser User-Agent string into browser, engine, OS, and device type via pattern matching against well-known UA token formats. Covers major desktop/mobile browsers and common search-engine/social crawlers.
parse_github_urlParse a github.com URL (repo, pull request, issue, commit, file, directory, or release tag) into structured components, including an optional #L10-L20 line-range anchor for file URLs. SSH remotes and other Git hosts are out of scope.
detect_redosStatically analyze a regex pattern for catastrophic-backtracking (ReDoS) risk. Flags nested unbounded quantifiers ((a+)+), overlapping alternation under a repeat ((a|ab)*), and adjacent overlapping quantifiers (\d+\d+). Heuristic linter, not a formal proof. Pass the pattern body only, no delimiters or flags.
analyze_regex_groupsParse a regex pattern's capture groups (numbered and named) and backreferences, and flag any backreference pointing at a group that doesn't exist or hasn't finished being defined yet at that point in the pattern (a forward reference or self-reference). Structural, non-AI check. Pass the pattern body only, no delimiters or flags.
build_cache_controlBuild an HTTP Cache-Control response-header value from a set of directives (public/private, no-cache, no-store, max-age, s-maxage, must-revalidate, immutable, stale-while-revalidate, etc.), with a plain-English explanation of each directive and warnings for combinations that cancel each other out. Pure string assembly, not validated against a live origin.
build_referrer_policyBuild an HTTP Referrer-Policy response-header value from one or more policy tokens (a fallback chain per the Referrer Policy spec), with a plain-English explanation of each token and warnings for risky choices like unsafe-url. Pure string assembly and lookup, not validated against live request traffic.
build_link_headerBuild an HTTP Link response-header value from one or more URL/rel pairs (preload, preconnect, prefetch, canonical, pagination, etc.), including the as/type/crossorigin attributes preload needs, with a plain-English explanation of each entry and warnings for common mistakes like a preload missing `as`. Pure string assembly and lookup; it never fetches or validates that a URL exists.
parse_ipv6_addressExpand an IPv6 address to its full 8-group form and compress it to its RFC 5952 canonical shortest form, classify its address type (loopback, link-local, unique-local, multicast, IPv4-mapped, documentation, or global unicast), and separate out an optional zone ID. Understands an embedded IPv4 suffix like "::ffff:192.168.1.1".
detect_filler_wordsScan prose for filler words (very, just, really), hedge phrases (kind of, I think), and clichés (at the end of the day, circle back), returning each matched phrase with its category, occurrence count, and character offsets, plus an overall density. Dictionary-based heuristic: a flagged word is a candidate to reconsider, not always wrong in context.
detect_wordy_phrasesScan text against a curated table of wordy/redundant phrases ("due to the fact that", "in order to", "at this point in time") and flag each occurrence with a suggested concise replacement, an occurrence count, and character offsets. Dictionary-based heuristic: a flagged phrase is a candidate to tighten, not always wrong in context.
check_oxford_comma_consistencyScan a document for comma-before-and/or usage in three-or-more-item lists and flag inconsistency: some lists using the Oxford (serial) comma while others in the same document don't. A heuristic over punctuation and sentence boundaries, not a full grammar parser.
calculate_percentage_changeCompute the percentage change from an original value to a new value as ((new - old) / |old|) * 100, plus the absolute difference, the direction (increase/decrease/no change), and the multiplier (new / old). The original value cannot be zero unless the new value is also zero.
resolve_json_pointerResolve an RFC 6901 JSON Pointer against a JSON document, in both the plain form ("" for the whole document, or "/foo/0") and the URI-fragment form ("#/foo/0"). Handles ~1 -> / and ~0 -> ~ escaping, and distinguishes an invalid document / malformed pointer from a syntactically valid pointer whose target is simply absent.
apply_json_merge_patchApply an RFC 7396 JSON Merge Patch document to a target JSON document and return the merged result. A null-valued key in the patch removes that key from the target; a patch that is not itself a JSON object replaces the target wholesale, which is how arrays are replaced rather than merged element by element.
convert_apr_apyConvert between a nominal annual rate (APR) and the effective annual yield (APY) for a given compounding frequency. With direction "apr-to-apy" the input is the nominal APR and the effective APY is computed as (1 + apr/n)^n - 1; with "apy-to-apr" the input is the effective APY and the APR is recovered as n * ((1 + apy)^(1/n) - 1). Both rates are percentages.
convert_iso8601_durationParse or format an ISO 8601 duration. With mode "parse", break a duration string like "P3DT4H30M" into its components and a total number of seconds (years and months use nominal 365-day / 30-day conversions and set an "approximate" flag). With mode "format", render a non-negative number of seconds into a canonical duration string using days/hours/minutes/seconds.
calculate_margin_markupSolve a pricing triangle: given exactly two of cost, price, gross margin percent, and markup percent, derive the other two plus the gross profit. Margin is profit as a share of the selling price (profit / price) while markup is profit as a share of the cost (profit / cost), so a 20% margin equals a 25% markup. Margin plus markup alone cannot be solved, since neither carries an absolute amount.
calculate_discountDiscount / markdown calculator. Give an original price plus exactly one of discountPercent, discountAmount, or finalPrice and the other two are derived. Pass discountPercents instead to stack discounts in sequence, where 20% then a further 10% is an effective 28% off rather than 30%. Pure arithmetic, no live price feed and no tax or shipping.
convert_subtitlesConvert a subtitle file between SubRip (.srt) and WebVTT (.vtt), optionally shifting every cue by a fixed number of seconds to resync it. The source format is auto-detected. Converting WebVTT to SRT drops cue identifiers, cue settings, and NOTE/STYLE/REGION blocks, since SubRip has no equivalent; every drop is reported as a warning.
calculate_break_evenFrom fixed costs, the variable cost per unit, and the selling price per unit, compute the break-even point: the unit count and revenue at which profit is exactly zero. The price must exceed the variable cost, since otherwise every unit sold loses money and break-even is never reached no matter the volume. Optionally pass a sales volume to also get the profit (or loss) at that volume and the margin of safety above break-even. Pure arithmetic, no live cost/pricing feed.
calculate_string_distanceCompute the Levenshtein edit distance between two strings: the minimum number of single-character insertions, deletions, and substitutions needed to turn a into b, plus a normalized 0-1 similarity ratio. Pass mode "damerau" to also count an adjacent-character transposition as a single edit (optimal-string-alignment variant, not unrestricted Damerau-Levenshtein).
calculate_video_bitrateCalculate the video file size from a target bitrate and duration, or the required bitrate to hit a target file size, with an optional separate audio bitrate. Pure arithmetic (file size bytes = total bitrate bits/sec * duration seconds / 8); no container parsing or encoding is performed.
calculate_rule_of_72Estimate how many years it takes an investment to double at a given annual rate (direction "rate-to-years"), or the rate needed to double in a given number of years (direction "years-to-rate"), using the Rule of 72 shortcut (72 / rate, or 72 / years). Also returns the exact result from the logarithmic doubling-time formula ln(2) / ln(1 + rate/100) plus the approximation error, since the Rule of 72 is most accurate for rates roughly between 6% and 10%.
convert_timecodeConvert between a frame count, SMPTE timecode (HH:MM:SS:FF, or HH:MM:SS;FF for drop-frame), and seconds at a given frame rate. Handles NTSC drop-frame timecode (29.97/59.94 fps) correctly via the standard SMPTE 12M algorithm, which skips frame numbers 00 and 01 at the start of every minute except every 10th minute so the nominal 30/60 fps digit count stays in sync with the true 29.97/59.94 fps signal. Pure arithmetic; no video file or container is read.
validate_geojsonValidate a GeoJSON document's structure against RFC 7946: the correct geometry type and coordinates shape for each of the seven geometry types, linear ring closure for Polygon/MultiPolygon, a Feature's required properties/geometry members, and a FeatureCollection's required features array. Also returns non-failing warnings when a Polygon ring winds the "wrong" way under RFC 7946's right-hand rule recommendation. Pure structural validation; no map is rendered and no tile server is contacted.
read_video_infoRead duration, resolution, and video/audio codec identifiers from an MP4 (ISO BMFF) or WebM (EBML/Matroska) file's container header. No frame decoding, demuxing, or transcoding is performed.
json_schema_to_typescriptConvert a JSON Schema document into TypeScript interfaces/types. Unlike json_to_typescript (which infers types from an example data payload), this reads the schema's declared type/enum/required keywords directly, so optional vs. required fields and enum/union types come through accurately. Understands local $ref into $defs/definitions, items for arrays, additionalProperties, and renders description as a JSDoc comment.
estimate_password_entropyEstimate a password's entropy in bits from the character classes it actually draws from (lowercase, uppercase, digits, symbols, other/Unicode), using the pool-size x length method. Returns a strength label and crack-time estimates at four guess-rate scenarios, plus warnings for repeated runs, sequential runs, short length, and single-character-class passwords. Measures resistance to brute force over the theoretical character space, not real-world guessability against a breach database.
calculate_emiCompute the Equated Monthly Installment (EMI) for a loan given principal, annual interest rate (as a percent), and tenure in years or months. Same reducing-balance math as calculate_loan, but reports a year-by-year principal/interest breakdown instead of a full monthly schedule.
parse_markdown_front_matterSplit a Markdown document into its YAML front matter (delimited by --- lines at the top) and body content, parsing simple scalar, list, and inline-array front matter values. Not a full YAML parser: nested mappings and multi-line block scalars are not supported.
generate_markdown_tocScan a Markdown document's ATX headings (# through ######) and build a nested table of contents with GitHub-style anchor slugs, disambiguating duplicate headings with -1, -2, ... suffixes. Setext-style headings are not recognized, and headings inside fenced code blocks are ignored.
align_markdown_tablesReformat every GitHub-Flavored-Markdown pipe table in a document so each column's pipes line up visually, padding cells to the widest content in that column and rebuilding the delimiter row to match, while preserving each column's declared alignment. Tables inside fenced code blocks are left untouched.
html_to_jsxConvert an HTML snippet to React JSX: class to className, for to htmlFor (and 60+ other attribute renames, SVG attributes like stroke-width to strokeWidth), inline style strings to style objects, self-closing void elements, comments to {/* */}, escaped braces in text, input value/checked to defaultValue/defaultChecked, and inline event handlers wrapped in arrow functions. Multiple top-level nodes are wrapped in a fragment. Optionally wraps the result in an exported function component. Returns the JSX plus warnings about anything that needs a manual look.
svg_to_reactConvert SVG markup into a React component, like SVGR: attributes are camelCased (stroke-width to strokeWidth, xlink:href to xlinkHref, class to className), inline styles become style objects, {...props} is spread onto the root <svg>, and the result is wrapped in `const Name = (props) => (...)` with a default export (optionally TypeScript with SVGProps<SVGSVGElement>). By default it strips the XML prolog, comments, <metadata> and Inkscape/Sketch/Illustrator editor data. Options can set width/height to 1em (adding a viewBox when missing), remove them, or replace fill/stroke colors with currentColor. Returns the component code, the bare JSX and warnings. Does not optimize path data.
docker_run_to_composeConvert one or more `docker run` commands into a docker-compose.yml (Compose Specification, no obsolete version: key). Handles \ line continuations, quoting, combined short flags (-it, -p8080:80), and several commands (one service each). Maps ports, volumes and --mount (declaring named volumes), env vars, --env-file, restart, networks (as external) and network_mode, healthchecks, resource limits, ulimits, --gpus, logging, labels, capabilities and the command after the image. Returns the YAML, the service names and warnings for flags with no Compose equivalent (like --rm). Parses only; nothing is executed.
check_spf_dmarcCheck an SPF record (RFC 7208) and/or a DMARC record (RFC 9989, which replaced RFC 7489) pasted as text, e.g. from `dig TXT example.com` and `dig TXT _dmarc.example.com`. Explains every SPF term and DMARC tag, counts the SPF terms that need a DNS lookup (limit 10; includes add their own, which this offline check does not follow), and flags issues with a severity: +all or ?all, no all mechanism, unknown mechanisms, bad ip4/ip6 ranges, ptr, terms after all, p=none, pct below 100, sp=none, a missing rua, invalid tag values, and (when domain is given) report addresses on another domain that need an authorization record. Makes no DNS queries.
validate_markdown_heading_hierarchyScan a Markdown document's ATX headings and flag heading-structure issues: a level that skips one or more levels going deeper, more than one H1, an empty heading, a duplicate heading at the same level, or a document that doesn't start at H1. Setext-style headings are not recognized, and headings inside fenced code blocks are ignored.
lint_conventional_commitLint a commit message against the Conventional Commits v1.0.0 spec: header format (type(scope)!: description), blank-line separation before the body/footers, footer token format, and BREAKING CHANGE casing. Returns the parsed type/scope/description/body/footers plus a list of error and warning findings.
calculate_savings_goalStandard time-value-of-money arithmetic for an ordinary annuity (contributions at the end of each month, compounded monthly). Mode "time-to-goal" solves for how many months it takes to reach a savings goal given a fixed monthly contribution; mode "contribution-for-target" solves for the monthly contribution required to reach the goal in a given number of months. Pure closed-form formulas, no live rate data.
calculate_annuity_payoutStandard time-value-of-money annuity formula, solved in either direction. Mode "payout-from-principal" takes a lump sum and computes the fixed periodic payout a given number of periods at a given rate would produce; mode "principal-from-payout" takes a desired periodic payout and computes the lump sum required to fund it. Structure "ordinary" assumes payments at the end of each period, "due" at the start (worth one extra period of growth each). Pure closed-form formulas, no live rate data.
calculate_roiCompute return on investment as a percentage from an initial and final value, and optionally the annualized return given a holding period in years. Unlike the present-value or compound-interest calculators, which grow a sum at a known rate, this derives the realized rate from actual start/end values. Pure arithmetic, no live rate feed.
lookup_emojiSearch a curated table of several hundred common emoji by name, GitHub-style shortcode, or keyword (mode "search", the default), or look one up exactly by shortcode, character, or category (modes "shortcode", "char", "category"). Not the full 3,000+ Unicode emoji set.
check_set_cookie_headerParse a Set-Cookie header value and validate its attribute combination: SameSite=None requires Secure, the __Host-/__Secure- name prefixes each require their own attributes (__Host- additionally forbids Domain and requires Path=/), a missing Secure attribute is flagged, and a Max-Age beyond 400 days is flagged since Chromium-based browsers clamp it to that cap.
test_gitignore_patternsTest a list of paths against a .gitignore file's contents, following git's matching rules: a leading `!` negates a pattern, a trailing `/` restricts a pattern to directories, and a pattern with no interior `/` can match at any depth. The last matching pattern wins. Does not replicate git's rule that a negation cannot re-include a file whose parent directory was already excluded.
compare_unit_pricesCompare two or more package options by price and quantity to find the true cost per unit (price / quantity) for each, flag the cheapest, and report how much more expensive each other option is by percentage. Unit-agnostic: pass quantity in whatever unit makes the options comparable (grams, litres, item count, etc.), no conversion is done.
csv_to_sql_insertPaste CSV data and get parameterless SQL INSERT statements back, with basic type inference: unquoted for integers/decimals and true/false, NULL for empty cells, single-quoted (with '' escaping) for everything else. Column and table identifiers are quoted per the chosen dialect so reserved words and mixed-case names round-trip safely.
generate_sri_hashCompute a Subresource Integrity (SRI) hash for a script or stylesheet's content: digest it with SHA-256, SHA-384, and SHA-512 and return each as the `<algorithm>-<base64>` string an HTML `integrity` attribute expects. Returns all three algorithms at once.
analyze_email_headersParse a raw email header block (e.g. Gmail's "Show original", or the headers from a .eml file) into structured fields, the Received hop chain in the order they appear, and any SPF/DKIM/DMARC verdicts already present in an Authentication-Results header. Never performs a live DNS/SPF/DKIM check, it only reports what the pasted headers already claim.
calculate_cash_breakdownBreak a cash amount down into the minimum count of bills/coins from a set of denominations, using the standard greedy largest-first approach (optimal for canonical currency denomination sets like USD, EUR, and GBP). Any amount that can't be represented exactly with the given denominations is reported as a remainder rather than rounded away.
calculate_changeCompute the change owed for a payment/tender pair (amount tendered minus amount due) and break that change down into the minimum count of bills/coins from the given denominations.
generate_etagCompute a content-hash ETag for the given content: digest it with MD5, SHA-1, or SHA-256, hex-encode it, and quote it into the exact string an ETag response header expects, with an optional weak-validator (W/) prefix. Hashes whatever content you give it; does not replicate any specific web server's own size/modification-time-based ETag algorithm.
build_range_headerBuild an HTTP Range request-header value (RFC 7233) from one or more byte ranges: explicit start-end pairs, open-ended "start to end of file" ranges, or suffix "last N bytes" ranges. Pass totalSize to resolve open-ended/suffix byte lengths and get overlap warnings. Pure string assembly and arithmetic; never makes a network request.
parse_content_range_headerParse a Content-Range response header (e.g. "bytes 200-1000/67589" or "bytes */67589" for an unsatisfiable range) into its unit, byte bounds, and total resource size, so you can verify what a server actually sent back for a partial/resumable request.
find_csv_duplicate_rowsParse CSV/TSV data and group rows that share the same value across a chosen set of key columns, or the entire row if no key columns are given, so you can spot and clean up accidental duplicates before an import. Row numbers count data rows only (the header row, if any, is not counted).
infer_csv_column_typesParse CSV/TSV data and infer each column's data type (integer, float, boolean, ISO 8601 date, or string) by classifying every non-empty value and finding the tightest type that fits them all: all-integer values infer as integer, a mix of integers and decimals infers as float, and anything that doesn't fit a narrower type falls back to string. Empty cells mark a column nullable but don't affect its inferred type.
build_vary_headerBuild a Vary response-header value from a list of request-header names (e.g. Accept-Encoding, Accept-Language, Authorization), de-duplicating case-insensitively and attaching a plain-English caching note for each well-known header explaining what it does to shared caches.
parse_vary_headerParse an existing Vary header value (with or without a leading "Vary:" label) into its individual header names, each with a plain-English note on what it does to shared caching.
validate_manifestValidate a web app manifest.json against the W3C Web App Manifest spec's installability requirements: name/short_name, start_url, a recognized display mode, and an icons array covering a 512x512 icon and a maskable icon, plus recommended theme_color/background_color fields.
lint_tool_schemaLint an LLM tool/function-calling definition (OpenAI, Anthropic, or MCP format — a single definition, an array of them, or a {tools:[...]} wrapper) for mistakes that measurably hurt model accuracy: a missing or vague description, an invalid tool name, a non-object top-level parameter schema, properties missing descriptions, an empty enum, a "required" entry with no matching property, or duplicate tool names.
build_permissions_policyBuild a Permissions-Policy response-header value from a list of browser features (e.g. camera, microphone, geolocation) and their per-feature allowlist ("self", specific origins, "*" for all origins, or an empty allowlist to disable the feature entirely), attaching a plain-English note on what each feature restricts.
parse_permissions_policyParse an existing Permissions-Policy header value (with or without a leading "Permissions-Policy:" label) into its individual feature directives, each with a plain-English note on what it restricts.
calculate_payback_periodFrom an initial investment and a series of periodic cash flows, compute the simple payback period (the point where cumulative cash flow turns positive), plus the discounted payback period when a discount rate is supplied. Pure arithmetic over a running cumulative sum, no live rate feed.
calculate_cd_maturityCompute the maturity value of a certificate of deposit given a principal, a nominal annual interest rate (as a percent), a term in months, and a compounding frequency (periods per year, e.g. 1 = annually, 12 = monthly). Also reports the effective annual yield (APY) implied by the nominal rate and compounding frequency, and, when an early-withdrawal penalty in months of forfeited interest is given, the net payout if withdrawn before maturity.
build_client_hintsBuild an Accept-CH response-header value from a list of client hint names (e.g. Sec-CH-UA-Platform, Sec-CH-DPR) the server wants the browser to send on future requests, attaching a plain-English note on what each hint reveals. Pass a subset of those names as criticalHintNames to also build a Critical-CH header value, for hints the server needs on the very first request of a session.
parse_client_hint_headersParse raw request-header text, one "Header-Name: value" per line, into decoded client hint values: brand/version pairs for Sec-CH-UA-style headers, booleans for ?1/?0 values, unquoted strings, and numbers. Lines whose header name isn't a recognized client hint are returned separately under "unrecognized" rather than dropped.
calculate_depreciation_scheduleBuild a year-by-year depreciation schedule for a fixed asset from its cost, salvage value, and useful life in years, using straight-line, double-declining-balance, or sum-of-years-digits. The double-declining-balance method clamps at salvage value rather than auto-switching to straight-line partway through.
analyze_keyword_densityCount keyword or phrase frequency and density (occurrences per 100 words) across a block of text, the way an SEO keyword-density checker does. With the default ngramSize of 1, common English stopwords and words shorter than minWordLength are excluded; for ngramSize 2 or 3 (phrases), stopwords are kept since useful phrases often contain them.
build_content_dispositionBuild a Content-Disposition response-header value for a file download or inline display, given a disposition type (inline/attachment) and filename. Non-ASCII filenames get an RFC 5987 filename* extended parameter alongside an ASCII-sanitized filename fallback.
parse_content_dispositionParse a Content-Disposition header value into its disposition type, filename (decoding an RFC 5987 filename* parameter if present), and raw parameters.
inspect_gif_framesWalk a GIF87a/89a file's block structure (image descriptors, graphic control extensions, the NETSCAPE2.0 application extension) to report frame count, total animation duration, and loop count, without decoding any LZW-compressed pixel data.
read_midi_infoParse a Standard MIDI File's header and every track chunk to report format (0/1/2), tick division, initial tempo, time signature, a per-track event/note-on count, and total duration in seconds, without any MIDI playback or synthesis.
calculate_cash_roundingRound a cashless total to the nearest payable cash denomination, the way a till rounds a card-priced total to the nearest 5 cents (or whatever the smallest circulating coin is) when settling in physical cash, commonly called "Swedish rounding" or "cash rounding".
calculate_moving_averagesCompute simple moving average (SMA) and exponential moving average (EMA) series from a list of numbers and a window size, a foundational technical-indicator building block for price/volume data. No live market-data feed is involved: this operates only on numbers you supply.
calculate_markdown_task_list_progressParse a Markdown document's `- [ ]`/`- [x]` task list checkboxes and report completion percentage overall and per section, where a section is the nearest preceding ATX heading of any level.
calculate_gpaCompute a weighted GPA from a list of courses, each with a letter grade and credit hours, using the standard 4.0-capped scale (A+ counts the same as A).
validate_editorconfigValidate an .editorconfig file: parse it into its root flag and glob sections, and flag structural mistakes against the EditorConfig spec — unknown properties, invalid values for known properties (indent_style, end_of_line, charset, indent_size, tab_width, max_line_length), a root declaration outside the file's preamble, and duplicate properties within the same section.
validate_tsconfigValidate a tsconfig.json file: parse it as JSONC (// and block comments plus trailing commas are stripped, matching how tsc itself reads the file) and flag structural mistakes — unknown top-level keys, unknown compilerOptions keys, wrong types, and malformed files/include/exclude/references/paths.
calculate_credit_utilizationCompute per-account and overall credit utilization (balance / limit) from one or more credit accounts, plus a qualitative tier matching common credit-scoring guidance: under 10% is excellent, under 30% is good (the generally recommended ceiling), under 50% is fair, and 50%+ is poor. Overall utilization is total balance over total limit, not an average of the per-account percentages.
calculate_debt_to_income_ratioCompute the debt-to-income ratio (total monthly debt payments / gross monthly income) from one or more monthly debt items, plus a qualitative tier matching common mortgage-lending guidance: under 36% is healthy, under 43% is moderate (43% is the back-end ratio ceiling many mortgage lenders use), under 50% is high, and 50%+ is severe. This is a back-end DTI (all monthly debt payments over gross income), not a front-end housing-only ratio.
analyze_caption_speedParse an SRT or WebVTT file and report the words-per-minute reading speed implied by each cue's word count and duration, flagging cues that exceed common broadcast accessibility guidelines (default: ~180 WPM, ~20 characters/second). HTML-style cue tags like <i> are stripped before counting words and characters.
validate_package_exportsValidate a package.json "exports" field: flag common dual-package (ESM/CJS) misconfigurations — a "types" condition not listed first, a missing "default" fallback, a "require"/"import" pointing at the wrong file extension, mixing subpath and condition keys in the same object, and a subpath pattern with no matching "*" in its target.
build_x_content_type_options_headerBuild the X-Content-Type-Options header (only one valid value, "nosniff") with a plain-English explanation of the MIME-sniffing attack it prevents, optionally validating a pasted header value or assessing MIME-sniffing risk for a given Content-Type.
build_accept_languageBuild an Accept-Language request-header value from a list of locale/quality pairs, sorting by descending quality (ties keep their original order) and omitting ";q=1" for the default weight, per RFC 9110 section 12.5.4.
parse_accept_languageParse an existing Accept-Language header value (with or without a leading "Accept-Language:" label) into its ranked locale/quality pairs, sorted by descending quality per RFC 9110 section 12.5.4 (a missing ";q=" defaults to 1).
build_prefer_headerBuild a Prefer request-header value from a list of preferences, each with an optional value and optional parameters, per RFC 7240. A value or parameter value that is not a valid HTTP token is quoted automatically.
parse_prefer_headerParse an existing Prefer header value (with or without a leading "Prefer:" label) into its list of preferences, each with an optional value and optional parameters, per RFC 7240.
calculate_blended_loan_rateCompute the single weighted-average interest rate across a list of loan balances (e.g. multiple credit cards or student loans being weighed for consolidation), weighting each loan's rate by its share of the combined balance, plus the combined monthly and annual interest accrual.
generate_csp_nonceGenerate a cryptographically random, base64-encoded nonce for a Content-Security-Policy script-src or style-src directive, along with ready-to-paste header directive and <script>/<style nonce="..."> tag snippets. Generates a fresh, unique value on every call.
build_retry_after_headerBuild a Retry-After response header value from either a non-negative delay in seconds or a target date/time, per RFC 9110 Section 10.2.3. A date is formatted as an HTTP-date (IMF-fixdate, e.g. "Wed, 21 Oct 2026 07:28:00 GMT"), the only date format the header allows.
parse_retry_after_headerParse an existing Retry-After header value (with or without a leading "Retry-After:" label) into either its delay-in-seconds form or its HTTP-date form, per RFC 9110 Section 10.2.3.
calculate_freelancer_rateSolve for the minimum hourly rate a freelancer must charge to hit a target annual take-home income, given billable hours per week, weeks actually worked per year, annual business overhead, and an optional profit margin/buffer applied on top.
build_priority_headerBuild a Priority request/response-header value from an urgency (0-7, default 3) and/or an incremental flag, per RFC 9218. At least one of the two must be given.
parse_priority_headerParse an existing Priority header value (with or without a leading "Priority:" label) into its urgency and incremental fields, per RFC 9218. Missing fields fall back to their spec default (u=3, i=false).
lookup_close_codeLook up what a WebSocket close code means, per the RFC 6455 close code definitions and the IANA WebSocket Close Code Number Registry reserved ranges.
list_close_codesList every well-known WebSocket close code (1000-1015) defined directly by RFC 6455, in ascending order.
build_hsts_headerBuild a Strict-Transport-Security response-header value from a max-age (in seconds) plus optional includeSubDomains and preload directives, flagging combinations that will not qualify for browser HSTS preload lists (which require max-age >= 1 year, includeSubDomains, and preload all together).
parse_hsts_headerParse an existing Strict-Transport-Security header value (with or without a leading "Strict-Transport-Security:" label) into its max-age, includeSubDomains, and preload fields, flagging whether it qualifies for browser HSTS preload lists.
build_x_robots_tagBuild an X-Robots-Tag response-header value from flag directives (noindex, nofollow, noarchive, nosnippet, notranslate, noimageindex, none, all, indexifembedded) plus optional max-snippet, max-image-preview, max-video-preview, and unavailable_after value directives, per Google's documented robots directive syntax. Optionally scope the header to a single crawler with userAgent (e.g. "googlebot").
parse_x_robots_tagParse an existing X-Robots-Tag header value (with or without a leading "X-Robots-Tag:" label, and with or without a leading crawler user-agent token) into the same shape build_x_robots_tag returns.
lookup_websocket_opcodeLook up what a WebSocket frame opcode means per RFC 6455: 0x0-0x2 are data frames (continuation, text, binary), 0x8-0xA are control frames (close, ping, pong), and the rest of the 4-bit opcode space is reserved for future protocol revisions.
list_websocket_opcodesList every WebSocket frame opcode (0x0-0xF), the full 4-bit opcode space defined by RFC 6455, in ascending order.
plan_debt_payoffSimulate paying off a list of debts under both the snowball strategy (smallest balance first) and the avalanche strategy (highest interest rate first), given a fixed extra monthly payment stacked on top of everyone's minimum payment, and compare months to payoff and total interest between the two.
audit_dependency_rangesParse a package.json's dependencies, devDependencies, peerDependencies, and optionalDependencies fields, classify each version specifier's range type, and flag wildcard ranges, the non-reproducible "latest" tag, git/URL/file/link dependencies, and packages listed with conflicting or redundant ranges across sections.
calculate_loan_refinanceCompare a current loan's remaining payments against a new refinanced loan on the same outstanding balance, and compute the monthly savings and the break-even month where cumulative monthly savings recoup the closing costs.
parse_multipart_form_dataParse a raw multipart/form-data request body into its individual parts: field name, filename (for file fields), headers, and decoded value, per RFC 7578. Give the boundary directly, or the request's Content-Type header value to extract it from automatically.
build_robots_meta_tagBuild a <meta name="robots" content="..."> tag from flag directives (noindex, nofollow, noarchive, nosnippet, notranslate, noimageindex, none, all, indexifembedded) plus optional max-snippet, max-image-preview, max-video-preview, and unavailable_after value directives, the exact same directive vocabulary as the X-Robots-Tag header. Scope the tag to a single crawler with name (e.g. "googlebot") instead of the default "robots".
parse_robots_meta_tagParse an existing <meta name="..." content="..."> robots tag (or a bare content attribute value on its own) into the same shape build_robots_meta_tag returns.
extract_svg_sprite_symbolsParse an SVG sprite sheet's <symbol> definitions and list each one's id, viewBox, width, and height, alongside a ready-to-paste <use href="#id"></use> usage snippet. A symbol with no id is skipped (counted in skippedCount) since it can't be referenced by a <use> element at all.
parse_commit_trailersParse a commit message's trailer block (Co-authored-by, Signed-off-by, Fixes, Reviewed-by, etc.) per git's RFC 822-style trailer convention (git-interpret-trailers): the last paragraph of the message, where every line is either a "Key: value" pair or an indented continuation of the previous trailer's value. Returns the trailers plus the message body with the trailer block removed.
extract_markdown_footnotesExtract every Markdown-Extra-style footnote reference ([^label]) and definition ([^label]: text, with 4-space/tab-indented continuation lines) from a document, then flag references with no matching definition and definitions that are never referenced.
calculate_tax_equivalent_yieldConvert between a tax-free municipal bond yield and the equivalent taxable-bond yield given a marginal federal tax rate (plus an optional state rate, summed into a combined rate). With direction "tax-free-to-taxable", taxableYield = taxFreeYield / (1 - combinedRate); with "taxable-to-tax-free", taxFreeYield = taxableYield * (1 - combinedRate). Pure formula, no live bond pricing.
generate_hotpGenerate an HOTP code (RFC 4226) for a given counter value from a base32-encoded shared secret. Supports 6-8 digit codes and SHA-1/SHA-256/SHA-512 (SHA-1 is the default per the RFC). The secret is never stored, only used for this one HMAC computation.
generate_totpGenerate the current TOTP code (RFC 6238) from a base32-encoded shared secret, plus how many seconds remain before it rotates. Defaults (30s period, 6 digits, SHA-1) match what Google Authenticator and most other authenticator apps use. Pass an explicit timestamp for deterministic/testable output.
validate_vinValidate a 17-character VIN via the ISO 3779 / NHTSA weighted mod-11 check digit at position 9, and split it into its World Manufacturer Identifier (WMI), Vehicle Descriptor Section (VDS), and Vehicle Identifier Section (VIS). Does not decode make, model, or model year.
validate_isinValidate a 12-character ISIN (ISO 6166) via its Luhn-style mod-10 check digit. The 2-letter country code and 9-character alphanumeric security identifier are transliterated (A=10 ... Z=35) into a numeric string before the Luhn algorithm is applied, the same way credit card numbers are checksummed.
calculate_mortgage_pointsCompare a loan with no discount points against the same loan at a reduced rate bought with points costing a percentage of the loan amount, and compute the monthly savings and the break-even month where cumulative savings recoup the points cost.
calculate_cookie_size_budgetParse a Cookie request header's name=value pairs and compute each cookie's serialized byte size and the total header size, flagging any cookie over RFC 6265's 4096-byte per-cookie guarantee and the total against a configurable budget (4096 bytes by default).
score_security_headersParse a pasted block of raw HTTP response headers (one "Name: value" pair per line) and score it against commonly recommended security headers (CSP, HSTS, X-Content-Type-Options, X-Frame-Options, Referrer-Policy, Permissions-Policy, and the three Cross-Origin-* isolation headers), with an overall 0-100 score and A-F letter grade. Evaluates the header text you give it, not a live request.
calculate_loan_prepayment_penaltyCompute the dollar penalty cost of paying a loan off early, given its remaining payoff balance and a prepayment penalty structure: either a flat percent of the payoff balance, or a fixed number of months of interest at the loan's annual rate.
detect_negative_amortizationGiven a loan's balance, annual rate, and a proposed monthly payment, check whether that payment covers the interest accruing each period. If not, report the shortfall and project how the balance grows instead of shrinking.
validate_ariaParse an HTML snippet and check every role and aria-* attribute against a curated table of WAI-ARIA 1.2 roles: unknown role names, an aria-* attribute not permitted on the given role, and a widget role missing its required aria-* state (e.g. role="checkbox" without aria-checked). Covers a curated subset of common roles, not the full ARIA spec.
verify_webhook_signatureRecompute an HMAC over a raw webhook payload and compare it to a provided signature, the check every webhook receiver (Stripe, GitHub, Shopify, etc.) has to do before trusting a request. Accepts a signature with or without the common "sha256=" prefix style, and matches against either hex or base64 encoding since providers differ on which they send.
estimate_data_transfer_timeEstimate the theoretical time to transfer a file of a given size over a connection of a given speed (decimal/SI file size units, standard networking speed units where Mbps is megabits/second and MBps is megabytes/second). Pass targetSeconds instead of speed/speedUnit to run the reverse calculation: the bandwidth required to hit that target time.
diff_json_patchCompare two JSON documents and generate an RFC 6902 JSON Patch (a list of add/remove/replace operations) that transforms the first into the second. Object keys produce add/remove/replace; arrays are compared position by position (not a minimal edit-distance diff, so a reordering is not detected as a move).
apply_json_patchApply a sequence of RFC 6902 JSON Patch operations (add, remove, replace, move, copy, test) to a target JSON document and return the result, or an error naming the first operation that failed and why.
parse_connection_stringParse a database connection URI (postgres://, mysql://, mongodb://, mongodb+srv://, redis://, amqp://, and similar generic scheme:// connection strings) into its scheme, credentials, one or more hosts (mongodb-style comma-separated host lists included), database/path, and query parameters. Pure offline parse: never connects, resolves DNS, or validates the driver/scheme combination.
validate_fence_language_tagsScan a Markdown document's fenced code blocks (``` or ~~~) and flag ones with a missing, unknown, or inconsistently-cased language tag, checked against a curated list of common syntax-highlighter language identifiers.
count_sql_placeholdersCount and list positional (?, $1) and named (:name, @name) parameter placeholders in a SQL query string, flagging a mix of placeholder styles in the same query and, for numbered placeholders, a gap or duplicate in the sequence. String literals and comments are skipped.
extract_markdown_link_referencesExtract every reference-style Markdown link ([text][ref], including the shortcut form [text][] and image references ![alt][ref]) alongside its [ref]: url "title" definition, then flag usages with no matching definition and definitions that no usage ever references.
lint_dockerfileLint a Dockerfile for 4 common mistakes: a base image with no tag or pinned to `:latest`, a final stage with no WORKDIR, a final stage with no USER (running as root), and ADD used on a plain local file/directory where COPY would do. Returns a list of findings plus the parsed build stages.
validate_mermaid_syntaxHeuristic syntax validation for a Mermaid diagram definition: recognizes flowchart/graph, sequenceDiagram, classDiagram, stateDiagram(-v2), erDiagram, and pie by their header keyword, then runs diagram-specific structural checks (unterminated subgraph, unbalanced brackets, malformed edges for flowcharts; a participant referenced before its explicit declaration for sequence diagrams). A line-based heuristic, not a full Mermaid grammar; never renders the diagram.
find_cron_overlapsGiven two or more standard 5-field cron expressions, find every minute over the next N days where two or more of them fire at the same time, for example a nightly backup job colliding with a report job. Returns each colliding minute with the zero-based indexes of every expression that matched.
extract_prompt_template_variablesExtract every {{variable}}-style placeholder from a prompt template, reporting each unique variable's occurrence count and flagging ones used more than once. Optionally cross-check against a providedVars list to flag variables the template uses that are missing, and provided names the template never references.
calculate_credit_card_minimum_paymentSimulate paying only the minimum payment on a revolving credit card balance every month, where the minimum payment is whichever is larger of a percent of the current balance or a fixed dollar floor. Reports months and years to payoff, total interest, total paid, and the first vs. final month's payment amount, so the "minimum payment trap" becomes visible.
check_dash_usageScan text for likely dash misuse: a hyphen surrounded by spaces where an em dash is probably intended ("word - word"), a double hyphen ("--") left unconverted, and inconsistent spacing around em dashes across the same document (mixing "word — word" and "word—word"). Pure regex heuristic, not a grammar parser.
check_acronym_expansionScan text for acronyms (2-10 character all-caps tokens) and flag any that are used before being expanded on first mention, per standard technical writing style guides. Recognizes both expansion conventions: the full term followed by the acronym in parentheses, and the acronym followed by its expansion in parentheses. Pure regex heuristic, not a dictionary of real acronyms.
calculate_sales_commissionCompute a sales commission from either a flat percent rate applied to the whole sale amount, or a tiered (marginal) rate structure where each tier applies only to the slice of the sale amount within it, the same way progressive tax brackets work. An optional base salary is added to the commission to get a total payout. Pure arithmetic, no live data.
build_character_classAssemble a [...] regex character class from named building blocks (lowercase, uppercase, digits, whitespace), custom character ranges, and literal custom characters, with optional negation. Pure string assembly, it never runs the resulting pattern against any input.
intersect_semver_rangesDetermine whether two semver ranges overlap and, if so, the intersecting range. Supports the same caret/tilde/comparison-operator/exact-version syntax (combined with || or space-separated AND) as the semver tool.
check_markdown_internal_linksScan a Markdown document's ATX headings, compute the GitHub-style anchor slug for each, and check every [text](#slug) internal link against them, flagging links that point at a heading that doesn't exist (or was renamed).
build_clear_site_data_headerBuild a Clear-Site-Data response-header value from a list of directives (cache, cookies, storage, executionContexts, or * for all of them), flagging redundant directives alongside the wildcard.
parse_clear_site_data_headerParse an existing Clear-Site-Data header value (with or without a leading "Clear-Site-Data:" label, quoted or unquoted) into its individual directives.
summarize_csv_column_statsParse CSV/TSV data and compute per-column summary statistics: null and unique-value counts for every column, plus min/max/mean/median/standard deviation for any column where every non-empty value parses as a plain number. A column with any non-numeric non-empty value is treated as categorical, reporting the most frequent value and its count instead.
find_duplicate_lockfile_versionsParse an npm (package-lock.json), yarn (yarn.lock), or pnpm (pnpm-lock.yaml) lockfile and report every package installed at more than one resolved version, with a count of how many install locations resolve to each version. Format is auto-detected from content unless given explicitly.
summarize_git_diff_statParse the per-file lines and trailing summary line of `git diff --stat` or `git log --stat` output into structured per-file insertion/deletion counts plus overall totals. Once a diff is wide enough that git scales the bar down, the per-file split is a proportional estimate (flagged via approximate) rather than an exact count, though the overall totals stay exact.
build_alt_svc_headerBuild an Alt-Svc response-header value (RFC 7838) from one or more alternative services, each a protocol-id (e.g. "h3"), optional host, port, max-age in seconds, and an optional persist flag, attaching a plain-English note for each well-known protocol-id.
parse_alt_svc_headerParse an existing Alt-Svc header value (with or without a leading "Alt-Svc:" label), including the special "clear" value, into its individual alternative services with plain-English protocol notes.
read_png_chunksWalk a PNG file's chunk stream directly from its bytes to report every chunk's type, byte offset, data length, and whether its stored CRC-32 matches the data, plus the decoded IHDR header and a handful of common ancillary chunks (tEXt, pHYs, gAMA, tIME, acTL). No pixel decoding, and zTXt/iTXt text is listed but not zlib-inflated.
estimate_audio_file_sizeEstimate how large an audio file will be before recording or exporting it. Mode "pcm" computes the uncompressed size from sample rate, bit depth, and channel count, plus a canonical 44-byte WAV header. Mode "compressed" computes it from a target bitrate. Pure arithmetic: no file is read, written, or encoded.
validate_routing_numberValidate a 9-digit US bank routing number (ABA/RTN) via the weighted mod-10 checksum from the ABA's routing number specification. Also identifies the routing number's type and Federal Reserve district from its first two digits.
audit_markdown_image_alt_textScan a Markdown document's inline images (``) and flag alt text accessibility smells: missing/empty alt text, a generic placeholder word used alone ("image", "photo"), alt text that's just the image's own filename, and a redundant "image of"/"photo of" prefix. Only inline image syntax is scanned; reference-style images and raw HTML `<img>` tags are not detected.
calculate_position_sizeCompute a trade's position size from account size, risk percent per trade, entry price, and stop-loss price: the dollar risk divided by the per-unit stop distance, floored to a whole share/unit count. Also infers the trade direction (long/short) and, given an optional target price, the risk/reward ratio and whether the target sits on the correct side of entry.
calculate_dividend_yieldCompute a stock's dividend yield (annual dividend per share / share price) and, given trailing EPS, its payout ratio (dividend / EPS) and retention ratio (100 - payout ratio). A payout ratio over 100% is valid and reported as-is. Pure arithmetic, no live quote feed.
check_known_hosts_fingerprintsParse an ssh_known_hosts file (or a single `host keytype base64key` line) and compute each entry's SHA256 and MD5 fingerprint the same way `ssh-keygen -lf` does, decoding the key type and, for RSA, the modulus bit length, flagging a weak RSA key (<2048 bits) or a deprecated ssh-dss (DSA) key.
detect_weasel_wordsScan text for weasel words: vague hedging phrases that imply evidence or consensus without naming a checkable source ("studies show", "many experts believe", "it is widely known"). Dictionary-based heuristic over a curated phrase list, distinct from detect_filler_words' "hedge" category, which covers personal uncertainty ("maybe", "I think") rather than unattributed third-party authority.
check_proto_field_numbersParse a .proto file's message definitions and flag reused field numbers, gaps in the field-number sequence that aren't covered by a "reserved" or "extensions" statement, and field numbers outside the valid range (1 to 536,870,911, excluding 19000-19999, reserved for the protobuf implementation). Each message (including nested ones) gets its own independent field-number space, and oneof fields count toward their enclosing message.
read_quicktime_metadataRead duration, dimensions, and video/audio codec identifiers from a QuickTime .mov file's moov/mvhd/trak atoms, plus the mvhd creation/modification dates and any udta user-data metadata tags (title, artist, copyright, description, etc). No frame is ever decoded, demuxed, or transcoded.
build_caa_recordBuild a CAA (Certification Authority Authorization) DNS record (RFC 8659) from a tag plus either a domain (the CA authorized to issue, or null for a disallow-all record) or a raw value, with optional "; key=value" params and a critical flag.
parse_caa_recordParse a CAA DNS record's zone-file text representation (as returned by `dig CAA` or stored in a zone file), e.g. `0 issue "letsencrypt.org"`, into its flags, tag, authorized domain, and any "; key=value" parameters.
validate_markdown_blockquote_nestingWalk a Markdown document line by line and flag blockquote nesting that jumps by more than one level at once, e.g. going straight from `>` to `>>>` without a `>>` line in between, or a quote block whose first line already starts at a nesting level above 1.
Configuration
UTILIX_API_KEYSecretOptional. A free Utilix API key from https://www.utilix.tech/dashboard. Enables the 21 hosted AI tools; the 300 offline tools never need it.
Automated checks
Ran Nov 13, 2026Full report →Passed our automated checks. That is not a security audit: read the code and give the server the narrowest access it needs. How checks work
- Source repository linked and reachablePass
- Open source license (OSI)Pass
- Commit or release in the last yearPass
- Package name not close to a popular packagePass
- Remote endpoint uses HTTPSPass
- Destructive tools are labelledPass
- No hidden instructions to the modelPass
- No critical advisory in the package or its dependenciesPass
README
View on GitHub@utilix-tech/mcp
MCP server exposing 322+ developer utility tools to Claude, Cursor, VS Code Copilot, and any MCP-compatible AI.
utilix.tech · Docs · npm
What is this?
@utilix-tech/mcp is a Model Context Protocol server that gives Claude, Cursor, and other AI tools access to 300 deterministic developer utilities (JSON formatting, encoding, hashing, text diff, regex, token estimation, PII detection and much more) that run locally with no API key, plus 21 optional AI tools that run on Utilix's hosted AI with a free API key.
Installation
Claude Desktop
Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):
{
"mcpServers": {
"utilix": {
"command": "npx",
"args": ["-y", "@utilix-tech/mcp"]
}
}
}
Restart Claude Desktop. You'll see utilix in the tools list.
Cursor
Add to .cursor/mcp.json in your project root or ~/.cursor/mcp.json globally:
{
"mcpServers": {
"utilix": {
"command": "npx",
"args": ["-y", "@utilix-tech/mcp"]
}
}
}
VS Code (Copilot)
Add to .vscode/mcp.json:
{
"servers": {
"utilix": {
"type": "stdio",
"command": "npx",
"args": ["-y", "@utilix-tech/mcp"]
}
}
}
Windsurf / Other MCP clients
{
"mcpServers": {
"utilix": {
"command": "npx",
"args": ["-y", "@utilix-tech/mcp"]
}
}
}
Hosted server (no install)
The same tools are served at https://mcp.utilix.tech (Streamable HTTP) for clients that take a URL. Send your API key from utilix.tech/dashboard as Authorization: Bearer utx_live_.... Each tool call counts toward your plan's daily request limit.
- Cursor (
~/.cursor/mcp.json):{ "mcpServers": { "utilix": { "url": "https://mcp.utilix.tech", "headers": { "Authorization": "Bearer utx_live_..." } } } } - VS Code (
.vscode/mcp.json):{ "servers": { "utilix": { "type": "http", "url": "https://mcp.utilix.tech", "headers": { "Authorization": "Bearer utx_live_..." } } } } - Claude Code:
claude mcp add --transport http utilix https://mcp.utilix.tech --header "Authorization: Bearer utx_live_..." - Codex (
~/.codex/config.toml):[mcp_servers.utilix]withurl = "https://mcp.utilix.tech"andbearer_token_env_var = "UTILIX_API_KEY"
Tool inputs are sent to Utilix to run (Utilix records which tool ran, not what you sent). If nothing should leave your machine, use the local npx setup above. Claude.ai, ChatGPT and Claude Desktop's connector settings need OAuth sign-in, which isn't supported yet, so use the local setup there.
AI tools (optional, free API key)
21 tools run on Utilix's hosted AI (Claude): ai_generate_regex, ai_explain_regex, ai_generate_shell_command, ai_explain_shell_command, ai_explain_sql, ai_explain_error_message, ai_explain_stack_trace, ai_generate_commit_message, ai_generate_pr_description, ai_generate_docstring, ai_generate_test_cases, ai_generate_json_schema, ai_generate_cron, ai_explain_css, ai_generate_dockerfile, ai_explain_dockerfile, ai_explain_nginx_config, ai_explain_k8s_manifest, ai_explain_terraform, ai_explain_iam_policy and ai_generate_svg.
They need an API key from utilix.tech/dashboard (free to create). Pass it as the UTILIX_API_KEY environment variable:
{
"mcpServers": {
"utilix": {
"command": "npx",
"args": ["-y", "@utilix-tech/mcp"],
"env": { "UTILIX_API_KEY": "utx_live_..." }
}
}
}
With Claude Code: claude mcp add utilix -e UTILIX_API_KEY=utx_live_... -- npx -y @utilix-tech/mcp.
- Usage counts toward your Utilix account's monthly AI limits, shared with the website: free accounts get 3 runs a month per AI tool, Pro 20 to 50.
- What you pass to an AI tool is sent to the Utilix API (and from there to Anthropic's API) to generate the answer. The key is only sent to
api.utilix.tech. - Without a key, the AI tools reply with setup steps instead of failing. The 300 local tools never need a key.
Available Tools
AI Agent Utilities (28 tools)
Tools built for LLM pipelines — token management, context compression, extraction, RAG, and security.
| Tool | Description |
|---|---|
estimate_tokens |
Estimate token count and cost for any text |
trim_to_tokens |
Trim text to fit a token budget (start/end/middle) |
chunk_text |
Split text into overlapping chunks by token count |
compress_html |
Strip scripts/styles/comments to minimize HTML for LLM ingestion |
compress_markdown |
Collapse blank lines, strip frontmatter, trim whitespace |
compress_json |
Remove nulls/empty arrays/objects, optionally sort keys |
summarize_for_llm |
Extractive summarization to fit a token budget (no LLM needed) |
extract_urls |
Extract all URLs from text or HTML |
extract_json |
Pull JSON objects from anywhere in LLM output |
extract_tables |
Parse HTML tables into JSON arrays |
extract_entities |
NER-lite: emails, phones, IPs, dates, credit cards, IBANs |
rerank_chunks |
TF-IDF reranking — sort chunks by relevance to a query |
score_relevance |
Score a single text's relevance to a query (0–1 + grade) |
expand_query |
Add synonyms and stemmed variants to improve retrieval |
diff_json |
Structural diff of two JSON objects — shows added/removed/changed paths |
validate_json_schema |
Validate data against a JSON Schema (Draft-07) |
repair_json |
Fix common JSON syntax errors (trailing commas, unquoted keys) |
flatten_json |
Flatten nested JSON to dot-notation keys |
merge_json |
Deep-merge multiple JSON objects |
sanitize_html |
Remove dangerous tags/attributes for safe rendering |
deduplicate_lines |
Remove duplicate lines, optionally case-insensitive |
extract_keywords |
TF-IDF keyword extraction with scores and positions |
detect_pii |
Detect PII (emails, phones, SSNs, credit cards, IBANs...) |
redact_pii |
Redact detected PII with a custom replacement |
detect_secrets |
Find leaked API keys, tokens, passwords in text |
detect_prompt_injection |
Score text for prompt injection attempts (0–1) |
JSON (12 tools)
format_json · minify_json · diff_json · json_to_csv · csv_to_json · json_path · validate_json · yaml_to_json · json_to_yaml · json_to_typescript · json_to_go · json_to_python
Encoding (8 tools)
encode_base64 · decode_base64 · encode_url · decode_url · encode_html · decode_html · encode_base32 · decode_base32
Hashing (3 tools)
hash · hash_password · verify_password
Text (8 tools)
word_count · convert_case · slugify · lorem_ipsum · escape_string · diff_lines · html_to_markdown · line_ops
Time (5 tools)
from_unix · parse_cron · cron_next_runs · diff_dates · convert_timezone
Units (2 tools)
convert_bytes · px_to_units
Color (4 tools)
parse_color · check_contrast · generate_palette · generate_shades
CSS (2 tools)
generate_gradient · minify_css
Code (7 tools)
format_sql · format_html · test_regex · minify_js · format_graphql · parse_docker_image · parse_env
Generators (6 tools)
generate_uuid · generate_password · generate_ulid · generate_random_data · generate_qr
API & Network (5 tools)
decode_jwt · build_curl · parse_curl · curl_to_code · generate_cors
Data Formats (4 tools)
validate_yaml · toml_to_json · xml_to_json · parse_csv
Misc (3 tools)
optimize_svg · char_to_codepoint · number_to_words
Example Prompts
Once installed, you can ask Claude things like:
- "Trim this text to 2000 tokens using the utilix trim_to_tokens tool"
- "Check this user input for PII before I log it"
- "Rerank these 10 chunks by relevance to my query about pricing"
- "Extract all JSON objects from this LLM response"
- "Compress this HTML page to minimize tokens before sending it to GPT"
- "Diff these two JSON configs and show me what changed"
- "Validate this data against my JSON Schema"
- "Convert this YAML to JSON"
- "Generate a UUID v4"
- "What's the SHA-256 of this string?"
Run without installing
npx @utilix-tech/mcp
Links
- Web app: utilix.tech — try every tool in the browser
- REST API: api.utilix.tech — same tools over HTTP
- Node.js SDK: @utilix-tech/sdk
- Python SDK: utilix-sdk on PyPI
- Docs: docs.utilix.tech
License
MIT
Utilix MCP questions
How do I add Utilix MCP to Claude Code?
Run claude mcp add --env 'UTILIX_API_KEY=<YOUR_UTILIX_API_KEY>' utilix-mcp -- npx -y @utilix-tech/mcp in your terminal. Add --scope user to make it available in every project.
How do I add Utilix MCP to Cursor?
Open ~/.cursor/mcp.json and add the server under mcpServers, using the Cursor config shown in the install section above.
What tools does Utilix MCP provide?
It exposes 300 tools, including format_json, minify_json, diff_json, json_to_csv, csv_to_json.
Do I need an API key to use Utilix?
No, all utilities run locally without the need for an API key.
What kind of AI tools can use this server?
It is compatible with Claude, Cursor, VS Code Copilot, and any other MCP-compatible AI client.
Can it help with LLM token management?
Yes, it includes tools for estimating token counts, trimming text to fit budgets, and chunking text by token count.
Does it support JSON validation?
Yes, it can validate JSON against schemas, repair syntax errors, and perform structural diffs.