Tell Claude what to test. It runs the tools.

pentestMCP is an open-source MCP server that hands Claude and other AI agents 40+ real security tools (Nmap, Nuclei, SQLMap, ZAP, Active Directory attacks) inside an isolated Docker container. Plain English in, structured findings out.

claude mcp add --scope user pentest-mcp -- docker run --rm -i ramgameer/pentest-mcp:latest

Run it in your terminal with Docker running. Using Claude Desktop or VS Code? See other setups.

Runs in Docker Async scans
40+security tools exposed over MCP
1 commandto pull the image and connect Claude
108GitHub stars and counting
MITlicensed, fork it and make it yours

The toolkit you already use, now callable by an agent

Every tool is wrapped as an MCP tool with a typed schema, so the model knows what each one does and how to call it. Pick a category to see what's inside.

Built so the agent never gets stuck waiting

A pentest agent is only useful if long scans don't freeze the conversation and your laptop stays clean. These are the design decisions that matter.

Async launch and fetch for slow scans

Nmap, Nuclei, SQLMap and Gobuster run in the background. Claude launches a scan, gets a process ID right away, keeps working, and fetches results when they're ready. No timeouts, no hung chats.

Isolated by default

Everything runs in a Docker container as a non-root user. Scanners and exploit tooling stay off your host.

Concurrency control

A semaphore caps simultaneous heavy scans so a chatty agent can't melt your machine or the target.

Works with your client

Claude Desktop, Claude Code, VS Code Copilot Chat, or any custom agent that can launch a stdio MCP server.

Web and AD, not just ports

OWASP ZAP active scans and AJAX spider, plus Kerberoasting, BloodHound collection, DCSync and more for Active Directory work.

What happens when you ask

  1. 1

    You ask in plain English

    "Find subdomains for the target, then scan the live ones for known CVEs."

  2. 2

    Claude picks the tools

    It reads the tool list from pentestMCP and chooses what to run, with which arguments.

  3. 3

    Docker runs them

    pentestMCP executes the real command-line tools inside the container and returns the output.

  4. 4

    Claude explains the results

    Raw output becomes a readable summary, and the next step is one sentence away.

Other ways to connect it

You need Docker running. The image ships with the tools preinstalled, so there's nothing else to set up.

One command registers the server for your user.

$ claude mcp add --scope user pentest-mcp -- docker run --rm -i ramgameer/pentest-mcp:latest

Want to build from source? Clone the repo and run docker build. Full steps are in the README.

If it saved you an hour, star it.

pentestMCP is built and maintained in the open. A star takes ten seconds and does three real things:

  • Helps other pentesters find it when they search for MCP security tools
  • Tells the maintainer which direction is worth building next
  • Keeps the project free and open for everyone
108
developers have starred it
Star pentestMCP

Only test what you're authorized to test

pentestMCP is for education and authorized security assessments. Get explicit written permission before scanning any system you don't own. Unauthorized scanning is illegal, and you are responsible for how you use these tools.