mcpsec

Overview

mcpsec is an AI-driven security tool designed specifically for testing MCP servers. It combines static analysis with dynamic runtime testing to identify vulnerabilities such as SSRF, LFI, command injection, and other security flaws. The tool has discovered multiple critical CVEs in real MCP servers and provides comprehensive testing through scanners, fuzzers, and AI-powered payload generation. It supports multiple connection methods (stdio, HTTP) and integrates well with CI/CD pipelines through SARIF output.

Try asking AI

After installing, here are 7 things you can ask your AI assistant:

• scan

  Scan MCP servers for security vulnerabilities via stdio or HTTP

  Try:Scan my MCP server via stdioTry:Scan an HTTP MCP server with authentication
• audit

  Perform static analysis on MCP server source code

  Try:Audit my local MCP server implementationTry:Audit an MCP server from a GitHub repository
• fuzz

  Perform protocol fuzzing on MCP servers to discover crashes

  Try:Fuzz my MCP server with standard test casesTry:Fuzz with high intensity test cases
• sql

  Scan MCP servers for SQL injection vulnerabilities

  Try:Test my MCP server for SQL injectionTry:Scan a MySQL server for SQL injection
• chains

  Detect dangerous tool combinations in MCP servers

  Try:Check for dangerous tool chains in my serverTry:Find potentially harmful tool combinations
• exploit

  Start an interactive exploitation session for MCP servers

  Try:Start exploitation session for my vulnerable MCP serverTry:Interact with a known vulnerable server
• rogue-server

  Run a rogue MCP server for client-side testing

  Try:Start a rogue server on port 9999Try:Run a server with all attack vectors enabled
• info

  Enumerate the attack surface of an MCP server

  Try:Get information about my MCP server's capabilitiesTry:List all tools exposed by my server
• setup

  Configure AI provider settings for enhanced scanning

  Try:Set up AI provider for enhanced scanningTry:Configure OpenAI API for mcpsec
• prompt-injection

  Scan for prompt injection vulnerabilities in tool descriptions

  Try:Check my server for prompt injection in tool descriptionsTry:Scan for hidden instructions in tool prompts
• auth-audit

  Audit MCP servers for authentication issues

  Try:Check my server for authentication problemsTry:Audit my MCP server for dangerous tool combinations
• capability-escalation

  Detect undeclared capability abuse in MCP servers

  Try:Check for capability escalation in my serverTry:Scan for undeclared capability abuse

Note: Inferred from CLI command examples and scanner descriptions in the README. The server appears to expose these as MCP tools based on command usage patterns.

Comparable tools

Installation

# Basic installation
pip install mcpsec

# With AI-powered features
pip install mcpsec[ai]

# Via Nix
nix-shell   # basic
nix-shell --arg withAll true   # all optional deps

{
  "mcpServers": {
    "mcpsec": {
      "command": "python",
      "args": ["-m", "mcpsec", "stdio"]
    }
  }
}

FAQ

How is mcpsec different from other static analysis tools?

mcpsec focuses on dynamic testing of live servers, proving exploitability rather than just identifying potential vulnerabilities. It also provides AI-powered payload generation and comprehensive fuzzing capabilities.

Does mcpsec work with all MCP servers?

Yes, mcpsec works with any MCP server that implements the standard protocol, whether via stdio or HTTP. It supports authentication methods and can auto-discover local servers.

How do I report vulnerabilities I find with mcpsec?

mcpsec provides responsible disclosure options. Critical findings should be reported to the affected project maintainers first, with coordination for public disclosure after fixes are implemented.

Compare mcpsec with