mcp-audit
Offline security scanner for MCP server configurations across AI coding clients.
Tested on macOS and Linux
Use it
macOS arm64
Install via uv tool
uv tool install mcp-audit-scanner
Check that it works
mcp-audit --help 2>&1The output should contain “Scan MCP configurations for security issues.”.
First thing to try
mcp-audit discover 2>&1Run mcp-audit discover to find the MCP configurations on your machine.
Linux x86_64
install mcp-audit via uv tool install mcp-audit-scanner
uv tool install mcp-audit-scanner
Check that it works
mcp-audit --help 2>&1The output should contain “scan”.
First thing to try
mcp-audit version 2>&1Run mcp-audit discover to find the MCP configurations on your machine.
What it does
- Scan MCP server configs for security risksTested
How to use it
- Run
mcp-audit check --path <config file>on an MCP config such as mcp.json. - Read the grade and score (out of 100) in the output.
- Run
- Open an interactive attack graph dashboardTested
How to use it
- Run
mcp-audit dashboard --path <config file> --no-open --port <port>. - Open http://127.0.0.1:<port>/ in your browser to view the Attack Graph.
- Run
- Save approved MCP servers to a lock fileNot tested
- Find risky combinations across serversNot shown working
How far pickfit got
macOS arm64
- Installs
- Starts
- 2 of 4 features work
- Works offline
Linux x86_64
- Installs
- Starts
- 1 of 4 features work
- Works offline
Other systems
Good for
- scan detected MCP configs for security issues
- verify approved MCP servers have not drifted in CI
- export scan results to SARIF for GitHub Security tab
How pickfit tested it
Each run starts from a clean machine and follows the steps above. How pickfit verifies
macOS arm64
Its service answers requests
- Replayed from scratch
- Works offline
- Checked on 30 Oct 2026
- Version 0.25.1
Differences the test machine hit
- README: CLI help implies standard --version flag
In the test: use `mcp-audit --help` or `mcp-audit version`; `--version` errors with No such option - README: mcp-audit check / mcp-audit scan with no args
In the test: use --path /tmp/mcpdemo/mcp.json to scan a chosen file; bare scan/discover only check auto-discovered client paths - README: mcp-audit lock # Write mcp-lock.json
In the test: add --offline to avoid online package version lookup so replay works with network off - README: mcp-audit dashboard # Open interactive attack graph dashboard
In the test: use mcp-audit dashboard --path <config> --no-open --port <port> and query http://127.0.0.1:<port>/; server keeps running until stopped
Linux x86_64
Its main function works
- Replayed from scratch
- Works offline
- Checked on 30 Oct 2026
- Version 0.25.1
Differences the test machine hit
- README: `mcp-audit --version` flag
In the test: use `mcp-audit version` (subcommand) or `mcp-audit --help` - README: `mcp-audit scan` on a risky config exits non-zero
In the test: exit code 1 is the documented outcome when findings are detected, not a failure