mcp-policy-testkit

mcp-policy-testkit is a CI-first policy and security testing toolkit for MCP servers. It checks manifests, configuration files, live MCP metadata obtained through handshake, and local source artifacts for common security and quality failures before release.

• Configuration hygiene checks for secrets, environment exposure, unsafe path mappings, and dangerous commands
• Real MCP handshake support for stdio-configured servers and HTTP JSON-RPC endpoints
• Tool and prompt metadata analysis for naming clarity, schema quality, destructive disclosures, prompt injection, and tool poisoning
• Source scanning for command injection, dynamic execution, unsafe file access, and resource exhaustion patterns
• Reports in terminal, JSON, Markdown, and SARIF
• Rule registry with enable/disable controls and project-level config
• GitHub Actions examples for CI use

python -m pip install .

For development:

python -m pip install -e ".[dev]"

mcp-policy-testkit lint-config path/to/mcp.json --format terminal --fail-on high

mcp-policy-testkit test path/to/server-or-config --format json --output reports/scan.json

If the input config includes a runnable MCP server connection, the scanner will attempt a live MCP handshake:

{
  "mcpServers": {
    "local-server": {
      "command": "python",
      "args": ["server.py"]
    }
  }
}

You can also target an HTTP MCP endpoint directly:

mcp-policy-testkit test https://example.com/mcp --format terminal

mcp-policy-testkit report --input reports/scan.json --format sarif --output reports/scan.sarif

• config: hygiene and manifest safety
• tool_quality: tool naming, schema quality, and destructive disclosure
• safety: prompt injection, tool poisoning, shadowing, prompt meta