Performs scored audits of code, websites, APIs, containers, infrastructure, and cloud accounts for AI agents and CI/CD pipelines.
Also known as wezter96/openaudit, @openaudit/cli
- Stars
- 1
- Tools
- —
- Downloads, 30 days
- —
- Latest version
- —
- Last commit
- Nov 14, 2026
- License
- MIT
What is openaudit MCP?
openaudit performs scored audits of code, websites, APIs, containers, infrastructure, and cloud accounts for AI agents and CI/CD pipelines.
Summary generated from the project's README
About openaudit MCP server
Performing scored audits of code, websites, APIs, containers, and cloud accounts is what openaudit enables for developers and AI agents. It consolidates 40 built-in auditors and 53 open-source scanners into a single unified report that provides a 0–100 score and an A–F grade across ten distinct areas like security, accessibility, and performance. Users can run audits via the terminal using Bun, integrate them into CI/CD pipelines, or connect them to AI clients like Claude Code and Cursor through an MCP server.
The system is designed to be safe on hostile input by treating targets as untrusted and making tools that run project code opt-in. It handles diverse targets including mobile builds like APKs and IPAs, infrastructure as code such as Terraform and Kubernetes, and live cloud accounts via Prowler. Results are delivered in multiple formats including interactive HTML, JSON, SARIF, Markdown, and CSV, ensuring that findings from multiple scanners are deduplicated into single, actionable items.
What you can do with it
- 1Automated security and quality auditsA developer can run a comprehensive scan on a codebase to find leaked secrets, vulnerable dependencies, and linting errors.
- 2AI-driven accessibility and privacy reviewsAn AI agent can use specialized playbooks to perform judgment-heavy tasks like reviewing authorization or privacy data flows.
- 3Live website and API scanningYou can audit a live URL for SEO, performance, and security headers to ensure production readiness.
- 4Cloud infrastructure compliance checksTeams can verify their AWS or Azure configurations against industry benchmarks using integrated cloud tools.
Written from the project's README. Every use case is checked against a quote from it.
How to install openaudit MCP
Formats verified 10 Nov 2026Add to CursorAdd to GooseInstall openaudit MCP in Claude Code
$ claude mcp add openaudit -- npx -y @openaudit/cliInstall openaudit MCP in Cursor
~/.cursor/mcp.json
{
"mcpServers": {
"openaudit": {
"command": "npx",
"args": [
"-y",
"@openaudit/cli"
]
}
}
}Install openaudit MCP in VS Code
.vscode/mcp.json
{
"servers": {
"openaudit": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"@openaudit/cli"
]
}
}
}Install openaudit MCP in VS Code CLI
$ code --add-mcp '{"name":"openaudit","type":"stdio","command":"npx","args":["-y","@openaudit/cli"]}'Install openaudit MCP in Claude Desktop
claude_desktop_config.json
{
"mcpServers": {
"openaudit": {
"command": "npx",
"args": [
"-y",
"@openaudit/cli"
]
}
}
}Install openaudit MCP in Codex
$ codex mcp add openaudit -- npx -y @openaudit/cliInstall openaudit MCP in Codex config.toml
~/.codex/config.toml
[mcp_servers.openaudit] command = "npx" args = ["-y", "@openaudit/cli"]
Install openaudit MCP in Gemini CLI
$ gemini mcp add -s user openaudit npx -- -y @openaudit/cliInstall openaudit MCP in Gemini settings.json
~/.gemini/settings.json
{
"mcpServers": {
"openaudit": {
"command": "npx",
"args": [
"-y",
"@openaudit/cli"
]
}
}
}Automated checks
Ran Nov 14, 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 reachablewezter96/openaudit is public and reachablegithub.com/wezter96/openaudit ↗Pass
- Package names this repository as its source@openaudit/cli names wezter96/openaudit as sourcenpmjs.com/package/@openaudit/cli ↗Pass
- Build provenance (Sigstore attestation)Sigstore attestation for @openaudit/clinpmjs.com/package/@openaudit/cli/v/0.5.0#provenance ↗Pass
- Open source license (OSI)MITspdx.org/licenses/MIT.html ↗Pass
- Commit or release in the last yearLast commit or release 2026-11-07github.com/wezter96/openaudit ↗Pass
- OpenSSF ScorecardNot checked
- Package name not close to a popular packageNo popular package one edit away from @openaudit/cliPass
- No critical advisory in the package or its dependenciesNot checked
- No hidden instructions to the model0 tool descriptions and the README read; no override phrases, credential paths or hidden charactersPass
- Destructive tools are labelledNot checked
- Remote endpoint answers the MCP handshakeNot checked
README
View on GitHubopenaudit
One scored audit of everything you ship. Code, websites, running apps, APIs, containers, infrastructure, cloud accounts and mobile builds, from your terminal, your CI or your AI agent.
Open the interactive report: openaudit's audit of this repository, rebuilt on every push.
openaudit runs 40 built-in auditors and wraps 53 established open-source scanners, linters and type checkers when they're installed. Ten agent playbooks cover the judgment-heavy work no rule can do, such as authorisation review, keyboard and screen-reader journeys, and privacy data flows. Everything lands in one report with a 0–100 score and an A–F grade per area.
- One issue, one finding. When gitleaks, TruffleHog and the built-in secret scan all find the same key, you get one finding that says who confirmed it, scored once.
- A report people actually read. A self-contained interactive HTML page with an overview, a tab per area and an OWASP Top 10 / CWE Top 25 view. It also comes as JSON, SARIF, Markdown and CSV.
- Built for agents. An MCP server, a JSON agent mode and playbooks that Claude Code, Cursor or Codex can run with their own tools.
- Safe on hostile input. Targets are treated as untrusted. Tools that would run project code are opt-in, and nothing builds or installs what it scans.
- Free. Every wrapped tool is free and open source. The only cost is tokens, if you choose to use the AI features.
Quick start
bunx @openaudit/cli . # a codebase
bunx @openaudit/cli https://example.com # a live site (add --browser for Playwright + axe-core)
bunx @openaudit/cli ./app-release.apk # a mobile build (.apk, .aab, .ipa)
bunx @openaudit/cli image:nginx:1.27 # a container image
bunx @openaudit/cli cloud:aws --only cloud # a live cloud account (Prowler)
bunx @openaudit/cli . -f html -o report # write the interactive report to report/
bunx @openaudit/cli doctor # which optional scanners are installed
openaudit needs Bun 1.4 or newer. Scanners are optional: an auditor whose tool is missing is skipped with an install hint, and the rest of the audit runs normally. To run from source, see Contributing.
What it audits
Every finding belongs to one of ten areas, and each area gets its own score and grade.
| Area | What openaudit looks for |
|---|---|
| Security | Leaked secrets (also in git history), injection, XSS and other risky code, weak crypto, disabled TLS or JWT checks, security headers, TLS, DNS and email spoofing protection, exposed files and APIs, container, IaC and cloud misconfiguration, risky mobile manifests, and optional dynamic scanning |
| Accessibility | WCAG checks in JSX, Vue, Svelte, HTML and CSS; axe-core and pa11y on rendered pages; Android layouts; keyboard and screen-reader journeys walked by an agent |
| Performance | Page weight, caching, compression, Core Web Vitals, Lighthouse, image and bundle size, container image size, mobile app size and native libraries |
| SEO | Titles and descriptions, crawlability, canonical URLs, sitemaps, structured data, broken links |
| Privacy | Trackers and cookies set before consent, analytics, ad and session-replay SDKs, personal data written to logs, mobile permissions |
| Best practices | Repository hygiene (README, licence, security policy, CI, lockfiles), Dockerfile and IaC lint, GitHub Actions hardening |
| Reliability | Type errors, broken pages and redirect chains, missing Kubernetes probes and limits |
| Maintainability | Linting, strict typing settings, duplication, dead code, deep nesting and oversized files |
| Supply chain | Known-vulnerable dependencies, unpinned versions and actions, OpenSSF Scorecard, licences, SBOMs, container image CVEs |
| Compliance | Licence obligations, cookie consent, OWASP Top 10 and CWE Top 25 mapping, CIS cloud benchmarks |
Languages and ecosystems
| Language | Security | Dependencies | Types, lint and quality |
|---|---|---|---|
| JavaScript / TypeScript | built-in rules, Semgrep | npm, pnpm, Yarn and Bun lockfiles via OSV.dev; npm / pnpm / bun audit | tsc, oxlint (with its React, jsx-a11y and Next.js rules when the project uses them), Biome, ESLint*, knip*, strict tsconfig checks |
| Python | built-in rules, Semgrep, Bandit | requirements*.txt, poetry.lock; pip-audit |
ruff, mypy*, type-checker configuration |
| Go | built-in rules, Semgrep, gosec | go.mod; govulncheck* |
staticcheck, golangci-lint* |
| Rust | Semgrep | Cargo.lock; cargo-audit |
clippy*, [lints] and unsafe policy checks |
| Ruby | built-in rules, Semgrep, Brakeman (Rails) | Gemfile.lock |
|
| PHP | built-in rules, Semgrep | composer.lock |
PHPStan |
| Java / Kotlin | built-in rules, Semgrep, mobsfscan (Android) | Maven and Gradle via osv-scanner and Trivy | PMD (Java) |
| C# / .NET | built-in rules, Semgrep | NuGet via osv-scanner and Trivy | |
| Swift / Objective-C | mobsfscan (iOS) | ||
| Shell | ShellCheck, set -euo pipefail checks |
||
| HTML, CSS, JSX, Vue, Svelte | accessibility and performance rules |
* Opt-in: run it with --only <id>, e.g. --only code.clippy. The next section explains why some tools are opt-in.
Every language also gets secret scanning (built-in, gitleaks, TruffleHog), duplication checks (jscpd) and licence scanning. openaudit also reads these formats:
- Containers: Dockerfiles and Docker Compose.
- Infrastructure: Kubernetes manifests, Terraform, and GitHub Actions workflows.
- API definitions: OpenAPI / Swagger, AsyncAPI and GraphQL.
- Mobile builds: Android APK / AAB and iOS IPA.
What it checks, by target
Wrapped tools run when they're installed.
Codebase
- Built in: accessibility, AI-agent config (hidden instructions in AGENTS.md and Cursor/Copilot rules, unpinned or plaintext MCP servers, over-broad agent permissions), API specs (OpenAPI, GraphQL), change coupling and knowledge risk from git history, Code Health (a 1 to 10 maintainability score per file using CodeScene's rules: complex, long and nested functions, complex conditionals, duplication, with hotspots from git history), dependency licence policy, end-of-life runtimes and frameworks (endoflife.date), language strictness, lockfile integrity and install scripts, performance, privacy (tracking SDKs, PII in logs), quality, repo hygiene, secrets, static security (SAST), vulnerable dependencies (OSV.dev)
- Wrapped tools:
- Secrets: git history*, gitleaks, TruffleHog
- SAST: Bandit, Brakeman, gosec, mobsfscan, Semgrep
- Types and lint: Biome, clippy*, ESLint*, golangci-lint*, mypy*, oxlint, PHPStan, PMD, ruff, ShellCheck, staticcheck, tsc
- APIs: oasdiff (breaking changes, in diff mode), Spectral
- Dependencies: cargo-audit, govulncheck*, Grype*, npm/pnpm/bun audit, osv-scanner, pip-audit, Trivy
- Supply chain: GuardDog (malicious packages)*, OpenSSF Scorecard, Syft licences*
- Quality and licences: jscpd, knip*, ScanCode*
- AI review: deepsec*
URL
- Built in: accessibility, DNS and email (SPF, DMARC, CAA, MTA-STS, subdomain takeover), exposed APIs (OpenAPI, GraphQL introspection), exposed files, performance, privacy and trackers, reliability, security headers, SEO, TLS
- Wrapped tools: Lighthouse, OWASP ZAP baseline*, pa11y, sslyze, testssl.sh*, ZAP active scan* (allow-listed hosts only)
Running web app (--browser)
- Built in: axe-core in Chromium, console errors, Core Web Vitals, failed and mixed-content requests, trackers and cookies set before consent
Containers
- Built in: Compose, Dockerfile, image (root user, age, size, secrets)
- Wrapped tools: Dockle*, hadolint, Trivy
Infrastructure as code
- Built in: GitHub Actions, Kubernetes, Terraform
- Wrapped tools: actionlint, Checkov, KICS*, kube-linter, kubeconform, Kubescape*, TFLint, zizmor
Mobile
- Built in: Android manifest (incl. binary AXML in APKs), bundled SDK versions checked against OSV, embedded secrets, iOS Info.plist (incl. binary plists in IPAs), layout accessibility, size and performance
Cloud (cloud:aws, azure, gcp, kubernetes)
- Wrapped tools: Prowler*
* Opt-in, because it's slow, noisy, sends more traffic to the target, spends tokens, or runs project code. Select it with --only <id>, e.g. --only code.clippy or --only agent.deepsec.
The built-in hidden-Unicode check (code.unicode) flags Trojan Source bidirectional controls, zero-width and other invisible characters, hidden prompt text in Unicode tag characters, variation-selector payloads and look-alike (homoglyph) identifiers in code, docs and agent instruction files such as AGENTS.md and Cursor rules.
The opt-in code.dependency-health auditor looks your pinned dependencies up on deps.dev and flags deprecated packages, yanked releases, dependencies whose source repository is archived, and direct dependencies with no release in years.
bunx @openaudit/cli list shows every auditor and playbook. The auditor catalogue lists them by target and area.
The report
Every run produces one report, in whichever formats you ask for: -f terminal,json,sarif,markdown,html.
- Score and grade. Each area scores 0–100 from the severity of its findings; info findings never lower a score. The overall score is the average of the areas that were actually examined. Skipped auditors are listed with the reason, so a clean result never hides a check that didn't run.
- Interactive HTML. One self-contained file with no external requests, so it opens offline and can be attached to a CI run or an email. It has an overview, a tab per area with filters and search, an auditors tab, and OWASP Top 10 / CWE Top 25 groupings. It works with JavaScript off, prints cleanly, and lets readers download the page itself or the same audit as JSON, SARIF, Markdown or CSV. Shared reports name the project folder, never your local path.
- Stable fingerprints. Each finding has a fingerprint that baselines, suppressions and
compare_reportsrely on. - Diff mode.
--diff origin/mainkeeps only findings in files changed since that ref, for pull requests. OpenAPI specs are also checked for breaking changes against that ref with oasdiff.
With AI agents
Some audits need judgment rather than patterns. openaudit is built so agents can drive it end to end:
- MCP server.
claude mcp add openaudit -- bunx @openaudit/cli mcp(or the equivalentmcp.jsonin Cursor or VS Code) gives an agentrun_audit,get_finding,compare_reportsand more, plus the playbook toolslist_playbooks,get_playbookandsubmit_findings. Your agent does the review with its own tools, and its findings merge into the same scored report. No extra API key is needed. - Headless agentic auditors.
openaudit . --agenticruns the playbooks through the Claude Agent SDK. The agent gets read-only tools, confined to the project and the target origin. It usesANTHROPIC_API_KEYor an existing Claude Code login. - AI triage.
--aiturns the findings into a prioritised fix plan. Only finding metadata is sent, never file contents. It usesANTHROPIC_API_KEY, else your Claude Code login, so a Claude subscription is enough. - Agent mode CLI.
--jsonprints a single JSON report to stdout, andopenaudit schemagives its JSON Schema. - Skill.
skills/openaudit/SKILL.mdteaches an agent the loop: baseline, triage, playbooks, fix, then verify withcompare_reports. - Read-only reference over MCP.
https://openaudit.se/mcpanswers questions about the auditors, playbooks, docs and config schema without installing anything. It cannot run audits; the local server above does.
Details are in apps/cli/README.md (CLI, CI, MCP setup) and packages/agent/README.md (playbooks, data handling).
In CI
The repository is also a GitHub Action. On pull requests it reports only findings in changed files, uploads SARIF to code scanning, attaches the reports as an artifact and writes a job summary:
permissions:
contents: read
security-events: write
steps:
- uses: actions/checkout@v7
with:
fetch-depth: 0 # diff mode compares against the base branch
- uses: wezter96/openaudit@main # pin to a commit SHA in real use
with:
fail-on: high
Inputs: target, fail-on, diff (defaults to the PR base; none audits everything), only, exclude, config, upload-sarif, upload-report. Outputs: score, grade, report-dir.
Elsewhere, run the CLI: bunx @openaudit/cli . --diff origin/main -f html,sarif -o openaudit-report --fail-on high. --fail-on <severity> sets the exit code, and --baseline previous.json reports only findings that are new since an earlier run. This repository audits itself in ci.yml and publishes the result to openaudit.se/report with site.yml.
Configuration
bunx @openaudit/cli init looks at the project and writes a commented openaudit.config.jsonc: a $schema for editor completion, failOn, suppressions for the test fixtures it finds, and the opt-in auditors that fit the stack. Suppressions (ignore.rules, ignore.paths, ignore.fingerprints) can carry a reason and an until date, after which they stop applying and openaudit warns. Every report discloses what each config or baseline suppressed, and why. Per-auditor settings include a dependency licence policy (settings["code.licenses"] with deny and allow lists of SPDX ids). Code Health reads CodeScene's .codescene/code-health-rules.json (rule weights and thresholds per path) and its @codescene(disable:"...") comments, so both tools score the same code the same way. The report says when a rules file applied.
A config found inside the audited target is written by whoever controls the target, so it can only narrow an audit or add its own licence policy. It cannot enable opt-in auditors, add credentials or switch off the --fail-on gate. Settings like that come only from flags or from a file passed with --config <path>. In CI, pass the repository's own config with --config only where it has been reviewed, such as on pushes to the default branch, never on pull requests. See apps/cli/README.md.
Safety
openaudit treats every target as hostile:
- Files: reads are confined to the project root and never follow symlinks out of it.
- Ignored files: what
.gitignoreexcludes isn't scanned. Wrapped tools that walk the folder get it as a skip list, and findings any tool still reports there are dropped and counted in the report. Committed files are always scanned, even if.gitignorematches them, so a repository can't hide its own code. History scans and the licence check read ignored files on purpose. - HTTP: user-supplied headers go only to the exact target origin, and credentials are stripped on cross-origin redirects.
- Agents: openaudit's own agents have no shell, and the browser is held to the target origin at the network layer. deepsec's reviewer does get a sandboxed shell, which is why it is opt-in.
- Wrapped tools: tools never build or install the target. They run from temporary directories so project config can't redirect them (pnpm hook files, pa11y config, oxlint JS plugins, Spectral custom functions). Linters whose config is code (ESLint, clippy, mypy, golangci-lint) are opt-in.
- Git: every git call overrides
core.fsmonitorand hooks, and diffs compare commits rather than the working tree. The secrets history scan runs on a fresh mirror clone. Code Health and change coupling read only commit metadata (git log --name-only) with signatures, pagers, mailmaps and external diff tools switched off. - Active scanning: the ZAP active scan runs only against hosts allow-listed in a trusted
--config. - Hosted URL-audit API: connections are pinned to DNS-validated public IPs.
- Evidence: secrets in evidence are redacted.
URL auditors send only low-volume GET/HEAD requests; the opt-in ZAP scans are the exception. Only audit systems you own or are authorised to test. Report vulnerabilities privately as described in SECURITY.md.
How it works
CLI · MCP server · HTTP API · GitHub Action
│
▼
target detection ─► engine: select auditors, run them in parallel, isolate failures
│
built-in · wrapped scanners · agent playbooks
│
▼
merge duplicates ─► score each area ─► terminal · JSON · SARIF · Markdown · HTML
| Path | What |
|---|---|
packages/core |
Report model (the public contract), engine, cross-tool correlation, diff mode, target detection, reporters |
packages/auditors |
Built-in and external-tool auditors: codebase, web, infra, mobile, cloud |
packages/agent |
Playbooks, agentic auditors (Claude Agent SDK), AI insights |
packages/mcp |
MCP server |
apps/cli |
The openaudit binary |
apps/server, packages/api |
Hono + oRPC HTTP API (/api-reference for OpenAPI docs) |
apps/web |
Website: landing page, live URL audit, report viewer, docs |
docs/ |
Images for the README and site, and a local architecture overview: bun docs/architecture.ts writes docs/architecture.html (gitignored) |
Contributing
Contributions are welcome: new auditors, better rules, fewer false positives, docs. CONTRIBUTING.md covers setup, the auditor contract, the security rules every wrapper must follow, and the pull request checklist. Please read the Code of Conduct first.
bun install
bun run test && bun run check-types && bunx biome ci .
bun run audit -- . # run the CLI from source
Community
openaudit is shared with OpenSverige, Sweden's open community for people building AI agents and MCP servers. Come show what you've audited, or tell us what openaudit missed.
License
MIT. The scanners openaudit wraps keep their own licences and are installed separately.
openaudit MCP questions
How do I add openaudit MCP to Claude Code?
Run claude mcp add openaudit -- npx -y @openaudit/cli in your terminal. Add --scope user to make it available in every project.
How do I add openaudit MCP to Cursor?
Open ~/.cursor/mcp.json and add the server under mcpServers, using the Cursor config shown in the install section above.
What can openaudit audit?
It audits code, websites, running apps, APIs, containers, infrastructure, cloud accounts, and mobile builds.
How does the scoring work?
Each area receives a score from 0–100 and an A–F grade based on the severity of findings; info findings do not lower the score.
Is it safe to run on untrusted targets?
Yes, targets are treated as untrusted, tools that run project code are opt-in, and nothing builds or installs what is scanned.