Skill convert
✓ VérifiéConvertissez des fichiers entre plus de 140 formats via une CLI ou un MCP server, avec une prise en charge des convertisseurs personnalisés créés par IA et de l'extraction de données de documents.
- Étoiles GitHub
- 8
- Fichiers
- 1
- Dernier commit
- 10 nov. 2026
- Licence
- MIT
- Outils autorisés
- —
Que fait la skill convert ?
Convertissez des fichiers entre plus de 140 formats via une CLI ou un MCP server, avec une prise en charge des convertisseurs personnalisés créés par IA et de l'extraction de données de documents.
Résumé généré à partir du SKILL.md de la skill.
À propos du skill convert
Cette compétence permet aux développeurs et aux agents IA de transformer des fichiers entre plus de 140 formats différents. Elle prend en charge une large gamme de médias, notamment des documents comme Word et PDF, des formats de données tels que JSON et Parquet, des images, de l'audio, de la vidéo, des e-books et même des captures d'écran de sites web. Les utilisateurs peuvent accéder à ces capacités via la CLI ctio pour les workflows en terminal ou via un MCP server hébergé pour les environnements sans accès shell.
Lorsque les convertisseurs standards ne répondent pas à des exigences spécifiques, la compétence permet de créer des convertisseurs personnalisés à l'aide de descriptions en langage naturel. Ce processus permet de construire un pipeline de transformation réutilisable qui peut être exécuté sur des structures de fichiers spécifiques sans envoyer de contenu sensible à une IA. Le système inclut également des outils spécialisés pour l'extraction de texte par OCR et l'extraction de données de documents via la fonctionnalité Parse. Les agents peuvent utiliser ces outils pour gérer des fichiers volumineux qui dépassent les limites de contexte local ou pour effectuer des transformations de données complexes en plusieurs étapes.
Ce qu'un agent peut faire avec
- 1Convertir des formats de donnéesTransformez un fichier JSON en feuille de calcul Excel pour une consultation plus facile.
- 2Construire des transformations personnaliséesCréez un convertisseur unique en décrivant une structure de données spécifique en langage naturel.
- 3Extraire des données structuréesExtrayez des champs spécifiques comme des totaux et des dates d'une facture PDF vers un format JSON.
- 4Capturer du contenu webTransformez l'URL d'un site web en direct en un document PDF.
Rédigé à partir du SKILL.md du skill. Chaque cas d'utilisation est vérifié par rapport à une citation de celui-ci.
Comment installer la skill convert
Chaque commande copie le dossier de la skill dans le répertoire des skills du client. Redémarrez le client ou démarrez une nouvelle session pour la charger.Installer la skill convert dans Plugin
/plugin marketplace add conversiontools/agent-skills /plugin install conversiontools@conversiontools-skills
Installer la skill convert dans Claude Code
~/.claude/skills/convert/SKILL.md
$ npx degit conversiontools/agent-skills/skills/convert ~/.claude/skills/convertInstaller la skill convert dans Claude Code (project)
.claude/skills/convert/SKILL.md
$ npx degit conversiontools/agent-skills/skills/convert .claude/skills/convertInstaller la skill convert dans Codex
~/.agents/skills/convert/SKILL.md
$ npx degit conversiontools/agent-skills/skills/convert ~/.agents/skills/convertInstaller la skill convert dans Gemini CLI
~/.gemini/skills/convert/SKILL.md
$ npx degit conversiontools/agent-skills/skills/convert ~/.gemini/skills/convertInstaller la skill convert dans Claude.ai
npx degit conversiontools/agent-skills/skills/convert convert && zip -r convert.zip convert # claude.ai → Customize → Skills → + → Create skill → Upload a skill → convert.zip
Vérifications automatisées
Exécuté le 16 nov. 2026Rapport complet →A réussi nos vérifications automatiques. Ce n'est pas un audit de sécurité : lisez les scripts avant de laisser un agent les exécuter. Fonctionnement des vérifications · Ce qu'il faut corriger pour gagner des points
- Dépôt source lié et accessibleconversiontools/agent-skills/skills/convert is public and reachablegithub.com/conversiontools/agent-skills/tree/HEAD/skills/con ↗Réussi
- Licence open source (OSI)MITspdx.org/licenses/MIT.html ↗Réussi
- Commit ou version publiée au cours de l'année écouléeLast commit 2026-11-10github.com/conversiontools/agent-skills/commits ↗Réussi
- Aucun téléchargement-exécution ni vol d'identifiants dans les scriptsNo scripts in the skill folderRéussi
- Aucune instruction cachée dans SKILL.mdNo override phrases, safeguard switches, credential requests or hidden charactersgithub.com/conversiontools/agent-skills/blob/HEAD/skills/con ↗Réussi
- Les outils préautorisés sont limitésNo pre-approved tools; the agent asks before using onegithub.com/conversiontools/agent-skills/blob/HEAD/skills/con ↗Réussi
- SKILL.md respecte le format Agent Skillsthe description does not say when to use the skillgithub.com/conversiontools/agent-skills/blob/HEAD/skills/con ↗Faille
SKILL.md
Voir sur GitHubConversion Tools File Conversion
Convert files between 140+ formats directly from your agent. Two surfaces are available: the ctio CLI (a single binary, preferred when the agent has shell access) and the MCP server (zero-install, works in any MCP-compatible client).
And when no standard converter fits, you can build a custom one from a plain-language description (see the next section).
Build a custom converter (AI Studio)
When no existing converter fits - a custom transformation, a specific output shape, a multi-step pipeline, or a file no standard converter handles - have one built for you from a plain-language description, instead of writing and running conversion code yourself. Describe what you want, Conversion Tools builds the converter, runs it on the file, and returns the result. The converter persists in the user's account, so a build started here continues at the web AI Studio.
Reach for this (instead of writing your own conversion code) when:
- the conversion needs tooling you don't have installed locally - every conversion engine is ready and sandboxed, nothing to install or break on your machine;
- the file is too large to load into context or onto the local machine;
- the file is sensitive and its contents shouldn't be sent to an AI - the converter is built from the file's structure, not its contents, and the conversion runs sandboxed;
- the output needs a specific shape, or the transformation takes multiple steps;
- the user wants a durable, reusable converter, re-runnable from the web, API, or here.
For a simple conversion an existing converter already covers, use convert_file / find_converter - don't build.
The flow (MCP tools, conversiontools: prefix)
studio_create_converter- create the converter and attach the input file. Returns aconverter_id+ a web link to continue in the browser.studio_chat- describe the transformation in plain language (e.g. "turn this CSV into JSON, one object per invoice with a nested line_items array"). Read the returnedoutcome:ask_user-> answer its question with anotherstudio_chatturn;ready/propose_workflow-> run it;refuse-> it only builds file converters.
studio_run- run the converter on the attached file (asynchronous).studio_run_status- poll untilSUCCESS(returns aresult_file_id) orERROR.studio_download_result- download the result. Always check the output actually matches the request before trusting it.
Building and chatting are free; only runs are metered on the user's plan.
Reuse a converter built earlier (don't rebuild)
Converters persist in the user's account, so one built last week is re-runnable today on a new file - no rebuild, same logic.
studio_list_converters- list the user's existing custom converters (name, id, input -> output, whether runnable). Check here first before building. Passsearchto filter.studio_get_converter- inspect one (what it does, what it's powered by, readiness).studio_attach_file- attach the new file to that converter, then pollstudio_get_converteruntil its status isidle(extraction settles), thenstudio_run->studio_run_status->studio_download_resultas above.
With shell access, the whole reuse loop is one command via the ctio CLI - ctio studio run <converter_id> <file> <output> - see "Reuse an AI Studio converter" under the ctio section below.
Choosing a surface
Prefer ctio when the agent can run shell commands (terminal coding agents like Claude Code, Grok Build, Cursor, and Codex). Reasons:
- Composable via Unix pipes (chain with
jq,grep,awk, or feed into anotherctiocall) - Streams stdin/stdout directly - no base64 encoding, no 5 MB size threshold, no temp files
- Handles arbitrarily large files (chunked streaming upload, no in-memory buffering)
- Faster on small files (no base64 overhead, snappy 500 ms default polling)
Use the MCP server when ctio is not installed, when running in a web-based assistant, or when the user explicitly asks for MCP.
Quick capability check:
which ctio || echo "ctio not installed - use MCP tools below"
ctio CLI
Full docs: https://conversiontools.io/docs/ctio. Releases: https://github.com/conversiontools/ctio/releases.
Install (one time)
Installing software is the user's call: if ctio is missing, ask before installing it, or use the MCP server instead. Install it only through a package manager:
Windows (scoop):
scoop bucket add conversiontools https://github.com/conversiontools/scoop-bucket
scoop install ctio
macOS / Linux (Homebrew):
brew install conversiontools/tap/ctio
Without scoop or Homebrew, point the user to the per-platform downloads at https://conversiontools.io/docs/ctio#installation rather than downloading binaries yourself.
Update: scoop update ctio or brew upgrade ctio.
Authenticate (one time)
ctio auth login
Paste an API token from https://conversiontools.io/profile. Stored at ~/.config/conversiontools/profiles.toml (Unix, 0600) or %APPDATA%\conversiontools\profiles.toml (Windows). Free tier: 100 conversions/month (10/day). Paid plans at https://conversiontools.io/pricing.
Convert (the verbs)
# File to file
ctio convert -t json_to_excel data.json out.xlsx
# stdin to file (no temp file, true streaming upload)
cat data.json | ctio convert -t json_to_excel - out.xlsx
# File to stdout (status JSON goes to stderr so the pipe stays clean)
ctio convert -t pdf_to_text invoice.pdf -
# Compose with other tools
ctio convert -t pdf_to_text invoice.pdf - | grep -i total
# Remote URL input (no local file needed)
ctio convert -t website_to_pdf --url https://example.com out.pdf
# Conversion options (repeatable)
ctio convert -t excel_to_xml in.xlsx out.xml --option header=yes --option show_columns=no
# Browse converters and their options
ctio list --from pdf --to excel --format pretty
ctio list --ai --format pretty --detail # AI-powered converters with option keys
ctio list --to parquet --format pretty
ctio describe xml_to_csv --format pretty # every option with its allowed values and default
# Inspect tasks
ctio task <task_id> --format pretty
ctio task <task_id> --wait --download out.xlsx
ctio task list --status ERROR --limit 10
When the API ignores an option or replaces an unsupported value with its default, ctio prints a warning: line on stderr and the conversion still runs. Treat a warning as "the result is not what was asked for": fix the option and rerun instead of handing the output to the user.
Text to speech and transcription
Text to speech, voice cloning and transcription have their own skill: voice (skills/voice/SKILL.md). Use it for narration, voice-overs, cloned voices and transcripts.
Reuse an AI Studio custom converter
If you or the user built a custom converter in AI Studio, ctio studio runs it on new files - no rebuild. (Building and chatting are MCP- or web-only; the CLI reuses converters that already exist.)
# List your custom converters to find a converter_id
ctio studio list
ctio studio list invoice # filter by name/description
# Run an existing converter on a NEW file, wait, and save the result
ctio studio run <converter_id> invoice-april.csv out.json
ctio studio run <converter_id> data.csv - # stream result to stdout
# Re-download the converter's most recent result
ctio studio download <converter_id> out.json
ctio studio run uploads the file, attaches it to the converter, waits for it to be ready, runs it, and downloads the result in one step (--wait is implied when an output path is given). To repeat the same converter across many files, loop it:
for f in invoices/*.csv; do ctio studio run <converter_id> "$f" "out/$(basename "$f").json"; done
Runs are metered on the user's plan; studio list is free.
Extract named fields from a document (Parse)
ctio parse (ctio 0.8.0 or later) sends a PDF, scan or photo to Parse and prints the extracted fields as JSON. It waits until the extraction finishes.
ctio parse invoice.pdf # free-form: Parse picks the fields
ctio parse invoice.pdf --schema fields.json -o out.json # only the fields you list
ctio parse invoice.pdf --schema-id <schema_id> # a schema saved in Parse
ctio parse --id <extraction_id> # fetch, or keep waiting for, an earlier extraction
- Key. Parse uses its own API key, not the Conversion Tools token. The user creates one at https://parse.conversiontools.io and sets
CT_PARSE_TOKEN; never put the key on the command line yourself. Without a key, use theparse_*MCP tools (see the parse-integration skill). - Fields file. A JSON array such as
[{"name": "total", "type": "number"}, {"name": "due_date", "type": "date"}], or an object with afieldsarray (the shape Parse exports). - Exit code. 0 = the data is on stdout. Non-zero = no data: a failed extraction prints the reason on stderr, and when
--timeout(default 1800 s) runs out ctio printsctio parse --id <id>to fetch it later. ctio 0.7.x and older printednullwith exit 0 instead of waiting: if you seenull, ask the user to update ctio. - Cost. One Parse page per document page; failed extractions are not billed. The same document with the same fields returns the stored result, and no pages are charged.
Conversion type names
The -t flag accepts the same conversion type names as the MCP convert_file tool's converter parameter (e.g. convert.pdf_to_excel or just pdf_to_excel - the convert. prefix is auto-prepended). Some converters also answer to a clearer name (ai_text_to_speech for ai_text_to_mp3, which also writes WAV). Run ctio list to browse the live registry and ctio describe <type> for a converter's options.
MCP Server (fallback path)
If ctio is not available, use the hosted Conversion Tools MCP server. If the MCP server is already connected, skip to Available Tools.
On claude.ai or in Cowork there is no shell, so the MCP server is the only path. This plugin bundles it, but each person connects it once: Customize > Plugins > Conversion Tools > Connectors > Connect, then signs in with their Conversion Tools account. If no conversiontools tools are available in such a session, tell the user exactly that, so they can connect it.
If the MCP server is not connected, add it to your agent. The server endpoint is:
https://mcp.conversiontools.io/mcp
That URL offers the conversion tools only, which keeps the server light on your context. Custom converters (studio_*) and document data extraction (parse_*) are opt-in toolsets: use https://mcp.conversiontools.io/mcp?toolsets=all (or e.g. ?toolsets=convert,studio). This plugin already connects with every toolset. If a studio_* or parse_* tool is missing, the server was added without them: tell the user to re-add it with ?toolsets=all.
Per-agent setup instructions (Claude Code, Grok Build, Cursor, OpenAI Codex, and any MCP client) are at https://conversiontools.io/docs/agents. Restart your agent after adding the server - MCP server changes require a restart to take effect.
On first use, you will be prompted to authenticate via OAuth in your browser. Free accounts get 100 conversions per month (10 per day). Paid plans available at conversiontools.io/pricing.
Available Tools
All tools use the conversiontools MCP server prefix: conversiontools:tool_name.
conversiontools:convert_file - Convert a file
The primary tool. The converter is auto-detected from the file extensions of input_path (or url) and output_path, or set explicitly with converter.
Give the input as exactly ONE of:
url: a public http(s) link to the file or web page - nothing to download or upload firstfile_id: fromconversiontools:request_upload_url, any sizefile_content: base64 of the file, 5MB at most (it costs context - prefer an upload for anything but tiny files)
Other parameters:
input_path(optional): Name of the input file, e.g.orders.xml- gives the input format and names the uploadoutput_path(optional): Name of the result, e.g.orders.xlsx- gives the output formatconverter(optional): Explicit converter type likeconvert.pdf_to_excel. Auto-detected if omitted.options(optional): Converter-specific options. Unsupported ones come back aswarnings.
The result has a download_url valid for about 30 seconds - download it right away. If status is running (long conversions such as video), call conversiontools:get_download_url with the task_id until it completes; do not start the conversion again.
conversiontools:get_download_url - Check a conversion / get a fresh link
Parameters:
task_id(required): Thetask_idreturned byconvert_file
Returns the current status and progress while running, and a fresh download link once done (use it when the first link expired).
conversiontools:request_upload_url - Upload large files
Get a one-time URL to upload a file of any size, so the bytes never pass through the conversation.
Parameters:
filename(required): Name of the file to upload
Flow for large files:
- Call
conversiontools:request_upload_urlwith the filename - Upload file to the returned URL using PUT
- Call
conversiontools:convert_filewith the returnedfile_id
conversiontools:list_converters - Browse available converters
Use this to discover what conversions are currently supported. New converters are added regularly - always check here for the latest.
Parameters:
category(optional):documents,data,images,pdf,audio,video,ebook,ocr,ai,subtitles,webinput_format(optional): Filter by input format, e.g.pdfoutput_format(optional): Filter by output format, e.g.csv
conversiontools:find_converter - Find the right converter
Parameters:
input_format(required): e.g.pdfoutput_format(required): e.g.excel
conversiontools:get_converter_info - Get converter details
Parameters:
converter(required): Converter type, e.g.convert.pdf_to_excel
conversiontools:auth_status - Check login status, and as which account
conversiontools:auth_login / conversiontools:auth_logout - Local server only
On the hosted server your agent handles sign-in itself (OAuth in the browser) and renews it when it expires, so these two tools are not offered there.
How to Convert Files
The hosted server cannot read local paths. Choose how to hand over the input; the response includes a download_url - download the result with curl right away.
Files already online
Pass the public link as url - nothing to upload:
conversiontools:convert_file({ url: "https://example.com/report.xml", output_path: "report.xlsx" })
Web pages work the same way, e.g. converter: "convert.website_to_pdf" with the page URL.
Small files (up to 5MB)
- Base64-encode the file
- Call
conversiontools:convert_filewithfile_content,input_path, andoutput_path - Download the result from the returned
download_url
# 1. Encode
base64_content=$(base64 -w 0 data.json)
# 2. Convert
conversiontools:convert_file({
input_path: "data.json",
output_path: "data.csv",
file_content: "<base64_content>"
})
# 3. Download
curl -sL "<download_url>" -o data.csv
Larger files (and the better choice for most local files)
- Call
conversiontools:request_upload_urlwith the filename to get a one-time upload URL andfile_id - Upload the raw bytes within 5 minutes:
curl -X PUT --upload-file ./orders.xml "<upload_url>" - Call
conversiontools:convert_filewith thefile_idinstead offile_content - Download the result from the returned
download_url(or pollget_download_urlifstatusisrunning)
Prompts
Hosts that support MCP prompts list convert, extract_data and build_converter as ready-made starting points (in Claude Code: /mcp__conversiontools__convert).
Supported Categories
205 converters across the categories below (snapshot, regenerated from the live registry). Use conversiontools:list_converters with a category, input_format, or output_format filter to discover specific converters - new formats are added regularly.
- Office documents (17) - Word (DOCX, DOC), PowerPoint (PPTX, PPT), Excel (XLSX, XLS), ODS, XML to/from PDF, HTML, CSV, JSON, ODS, Excel-XML/XLSX
- PDF (15) - PDF to/from Word, Excel, CSV, HTML, Text, JPG, PNG, SVG, TIFF; plus OXPS to PDF, PDF to XPS, Website to PDF
- Data formats - JSON family (10) - JSON, JSON Objects, XML, YAML to/from CSV, Excel, JSON, XML, YAML
- Data formats - CSV (6) - CSV from BSON, RSS, Sitemap, XML; CSV to JSON, XML
- Data formats - XML (2) - Validators and formatters
- Data formats - JSONL (4) - JSONL to/from CSV, Excel
- Data formats - Parquet (9) - Parquet to/from CSV, Excel, JSON, JSONL, XML
- Web feeds to Excel (3) - BSON, RSS, Sitemap to Excel
- Images (57) - PNG, JPG, WebP, AVIF, HEIC, JXL (JPEG XL), SVG, BMP, TIFF, GIF mutual conversion; plus output to PAM, PGM, PPM, YUV; plus EXIF removal
- E-books (18) - EPUB, MOBI, AZW, AZW3, FB2, FBZ, PDF mutual conversion
- Audio (5) - MP3, WAV, FLAC mutual conversion
- Video (4) - MP4, MOV, MKV, AVI to MP4; plus audio extraction to MP3
- Subtitles (19) - SRT, VTT, ASS, CSV, Excel (XLSX, XLS), Text bidirectional
- Markdown (8) - Markdown to/from PDF, DOCX, HTML, EPUB, LaTeX, Text
- HTML / Web (8) - URL/HTML to PDF, JPG, PNG, Word, CSV; HTML metadata extraction; HTML table to CSV; website screenshots
- OCR (6) - Extract text from JPG, PNG, scanned PDFs; output as Text or searchable PDF
- AI-powered (14) - see the expanded section below
AI-powered converters (expanded)
These are the highest-leverage converters in the catalog. When standard converters produce poor results, when the input is messy or unstructured, or when the user wants natural-language output, reach for these. Type names use the convert.ai_* prefix.
Document data extraction
Pull structured data out of complex documents and images. AI handles layout drift, tables across page breaks, handwritten fields, and mixed languages where pdf_to_csv / pdf_to_excel struggle.
| Type | What it does |
|---|---|
convert.ai_pdf_to_json |
Extract structured data from PDF as JSON. Best for invoices, receipts, forms, statements. |
convert.ai_pdf_to_csv |
Same as above but flattened to CSV rows. |
convert.ai_pdf_to_excel |
Same as above as an Excel workbook (preserves multi-table structure). |
convert.ai_pdf_to_markdown |
Convert PDF to Markdown with formatting preserved (headings, lists, tables). Good for converting reports/papers into LLM-friendly Markdown. |
convert.ai_jpg_to_json |
Extract structured data from a JPG (photographed receipts, screenshots of tables, etc.). |
convert.ai_png_to_json |
Same for PNG. |
Use when: standard pdf_to_csv/excel returns garbled output, the document has nested tables, OCR alone is not enough, or the user wants typed fields (dates, amounts, IDs) rather than raw text.
AI speech (text to speech, transcription)
Text to speech, voice cloning and transcription have their own skill: voice (skills/voice/SKILL.md). Use it for narration, voice-overs, cloned voices and transcripts.
AI Subtitle Translation
| Type | What it does |
|---|---|
convert.ai_translate_srt |
Translate an SRT subtitle file into another language while preserving timestamps. Specify the target language via the language option. |
Use when: user has subtitles in one language and needs them in another. Faster and cheaper than human translation, good enough for most viewing use cases.
Discovering Converters
When unsure if a specific conversion is supported, use the discovery tools rather than guessing:
# Check what PDF can convert to
conversiontools:list_converters({ input_format: "pdf" })
# Check what can produce Parquet
conversiontools:list_converters({ output_format: "parquet" })
# Browse all data converters
conversiontools:list_converters({ category: "data" })
# Find the right converter for a specific pair
conversiontools:find_converter({ input_format: "xlsx", output_format: "parquet" })
# Get full details and options for a converter
conversiontools:get_converter_info({ converter: "convert.csv_to_parquet" })
Tips
- Auto-detection works well - just provide file paths with correct extensions and skip the
converterparameter - Use
conversiontools:list_converterswith filters when unsure what's available - AI converters (
convert.ai_pdf_to_json, etc.) use AI to intelligently extract structured data from complex documents - use these when standard converters produce poor results - OCR converters are for scanned documents and images containing text - use when the source is an image or non-searchable PDF
- For batch conversions, upload a ZIP, 7z, XZ, or RAR archive containing all files - they will all be converted and returned as a
.result.zipfile. Alternatively, callconversiontools:convert_fileonce per file.
Fichiers de cette skill
1Questions sur la skill convert
Comment installer la skill convert dans Claude Code ?
Exécutez npx degit conversiontools/agent-skills/skills/convert ~/.claude/skills/convert. Claude Code charge les skills depuis ~/.claude/skills (tous les projets) ou .claude/skills (un seul projet) et utilise une skill lorsque sa description correspond à la tâche.
Quels clients peuvent utiliser la skill convert ?
Tout client qui lit le format Agent Skills : Claude Code, Codex, Gemini CLI et claude.ai (envoi sous forme de ZIP). Les scripts de la skill ne s'exécutent que si le client autorise l'exécution de code.
Comment puis-je convertir des fichiers si je n'ai pas d'accès shell ?
Vous pouvez utiliser le MCP server Conversion Tools hébergé, qui est une option sans installation fonctionnant dans n'importe quel client compatible MCP.
Puis-je créer un convertisseur pour une transformation qui n'est pas standard ?
Oui, vous pouvez construire un convertisseur personnalisé en utilisant des descriptions en langage naturel via les outils d'AI Studio.
Que se passe-t-il si je dois extraire des données spécifiques d'un PDF ?
Vous pouvez utiliser la commande ctio parse pour envoyer un PDF, un scan ou une photo à Parse et extraire des champs au format JSON.