<p align="center" <picture <source media="(prefers-color-scheme: dark)" srcset="https://raw.githubusercontent.com/yatuk/mcpradar/main/docs/logo-dark.svg" <source media="(prefers-color-scheme: light)" srcset="https://raw.githubusercontent.com/yatuk/mcpradar/main/docs/logo-light.svg" <img src="https://raw.githubusercontent.com/yatuk/mcpradar/main/docs/logo-light.svg" width="160" alt="MCPRadar" </picture </p
<h1 align="center"MCPRadar</h1
<p align="center" Security scanner for Model Context Protocol servers.<br Detect tool poisoning, prompt injection, unsafe configuration, vulnerable dependencies, and behavioral drift before they reach an agent. </p
<p align="center" <a href="https://github.com/yatuk/mcpradar/actions/workflows/ci.yml"<img src="https://github.com/yatuk/mcpradar/actions/workflows/ci.yml/badge.svg" alt="CI"</a <a href="https://pypi.org/project/mcpradar/"<img src="https://img.shields.io/pypi/v/mcpradar?color=blue" alt="PyPI version"</a <a href="https://pypi.org/project/mcpradar/"<img src="https://img.shields.io/pypi/pyversions/mcpradar?color=blueviolet" alt="Supported Python versions"</a <a href="LICENSE"<img src="https://img.shields.io/badge/license-MIT-green" alt="MIT License"</a </p
<p align="center" <a href="#quick-start"Quick start</a · <a href="#capabilities"Capabilities</a · <a href="#usage"Usage</a · <a href="#github-actions"GitHub Actions</a · <a href="#public-leaderboard"Leaderboard</a · <a href="docs/cli-reference.md"Documentation</a </p
Overview
MCP servers expose instructions, tools, schemas, resources, and privileged integrations to AI agents. Security problems in those surfaces are often invisible to conventional package or network scanners. MCPRadar inspects the protocol surface, source code, configuration, and software supply chain, then records snapshots so later scans can identify meaningful changes.
A 2025 study of 1,899 MCP servers reported general vulnerabilities in 7.2% of servers and MCP-specific tool poisoning in 5.5% (arXiv:2506.13538). MCPRadar is designed to make this class of risk reviewable in local development and CI.
Quick start
Scan a running HTTP MCP server without installing MCPRadar:
uvx mcpradar scan http://localhost:8080Install the CLI for repeated use:
uv tool install mcpradar
mcpradar --helpPython 3.11 or newer is required. pip install mcpradar is also supported.
Capabilities
| Area | What MCPRadar provides | |---|---| | MCP inspection | Cursor-aware enumeration of tools, prompts, resources, templates, and server instructions | | Detection | Versioned rule catalog for poisoning, injection, schema abuse, secret exposure, transport issues, and cross-server attack paths | | Source analysis | Python and JavaScript/TypeScript analysis for SSRF, unsafe deserialization, command and SQL injection, Trojan Source, and description-code inconsistency | | Configuration | Detection of poisoned MCP/agent configuration, hooks, over-broad permissions, and suspicious package names | | Supply chain | Package fetching without install scripts, OSV dependency checks, CycloneDX SBOM output, hashes, and provenance | | Change monitoring | SQLite snapshots, tool fingerprints, and cosmetic/behavioral/security diff classification | | CI and policy | JSON, SARIF, CEF, policy-as-code gates, suppressions, and deterministic signed snapshots | | Isolation | Disposable Docker/Podman sandbox for untrusted stdio servers with bounded CPU, memory, processes, and output |
The generated detection rule catalog is the authoritative list of rules and severity mappings.