Back to search
MCP-VulnLab
MCP serverGitHub
Automated static AST + dynamic harness vulnerability fuzzer for open-source MCP servers, LangGraph tools, and LLM plugins, plus custom NVIDIA garak red-teaming probes.
Safety scan
A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.
Badge for your README
Markdown
[](/listings/cuhawk--mcp-vulnlab)HTML
<a href="/listings/cuhawk--mcp-vulnlab"><img src="/api/listings/cuhawk--mcp-vulnlab/badge.svg" alt="findsafeskills safety rating" /></a>Install
i
What was scanned
The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.
README.md · README · 38 lines
- # MCP-VulnLab
- MCP-VulnLab reviews local MCP tool source and metadata and validates in-memory tool mocks.
- Install it with `pip install -e '.[test]'` on Python 3.12 or newer.
- ## Static source review
- Run `mcp-vulnlab . --sarif` to emit SARIF 2.1.0 or omit the flag for JSON advisories.
- The existing `--format sarif` option is also supported. Findings are review
- candidates. Python tracking is intraprocedural and conservative. The JS/TS
- pass matches source patterns in MCP modules without resolving data flow.
- The source auditor does not import or execute modules under review.
- ## Offline mock validation
- Use `await run_mock(tool, arguments, schema, path_fields=("path",))` from
- `mcp_vulnlab.schema_validator` with a trusted async mock that takes one
- argument dictionary. It applies a compact UTF-8 JSON byte budget, validates
- against JSON Schema Draft 2020-12, and normalizes selected top-level path
- fields within a virtual POSIX root. The timeout covers the callable and
- requires cooperative cancellation. Schema references resolve within the
- document. Path normalization performs no filesystem access.
- ## Tool metadata review
- Run `python -m mcp_vulnlab.garak_probes.mcp_tool_shadowing tools.json --baseline trusted.json`.
- Each file contains a JSON array of tool records with a `name` and optional
- `description`. The baseline is optional. The scorer flags instruction-override
- patterns, duplicate normalized names, and descriptions that differ from the
- trusted baseline. A score of 1 indicates a review candidate.
- To use the optional adapter with a garak installation that provides
- `always.Passthru`, place `mcp_tool_shadowing.py` in that installation's
- `garak/probes` directory and install MCP-VulnLab in the same Python
- environment. Select `mcp_tool_shadowing.MCPToolShadowing` and configure
- its `tools` and optional `baseline` lists under
- `plugins.probes.mcp_tool_shadowing.MCPToolShadowing`. The probe scores the
- configured metadata and records the results without calling the generator.
Ratings and comments
Sign in to rate this skill.
Sign inFree to use. If it’s helped, pass it on.