MCP server: One MCP server.

Darkmoon

@ascit31MCP server · External

Drive a self-hosted Darkmoon Pro AI pentest instance: start runs, read campaigns and findings.

Penetration testingv0.1.0GPL-3.0TypeScript
Freeno charge per call

Runs on your machine from its package; the steps are below.

What it does

  1. Reads
  2. Writes
  3. High impact

At most: moves money, runs code or commands, deletes data, or sends messages for you.

What it costs

Free

Odel charges nothing per call. The publisher may charge for its own service.

Runs locally over stdio

This server isn't hosted — your MCP client launches it from a package registry. Use one of the commands below, or drop the config into your client (e.g. Claude Desktop).

npm@darkmoon_ai/mcp-serverv0.1.0

Run

npx -y @darkmoon_ai/mcp-server

MCP client config

{
  "mcpServers": {
    "darkmoon": {
      "command": "npx",
      "args": [
        "-y",
        "@darkmoon_ai/mcp-server"
      ]
    }
  }
}
README2 min read

@darkmoon_ai/mcp-server

A Model Context Protocol server that lets an MCP client (Claude Desktop, Goose, Continue, LibreChat, ...) drive Darkmoon, an open source (GPL-3.0) autonomous AI penetration testing platform.

Requires Darkmoon Pro

The Darkmoon engine and CLI are open source. This server talks to the Darkmoon Dashboard API, which is part of Darkmoon Pro and always self-hosted: there is no public hosted endpoint, so you supply the base URL of your own instance. It does not work against the open source CLI alone.

Tools

ToolDescription
run_pentestStart an autonomous pentest against one authorized target and return the run_id
get_run_statusReport running, completed, error or unknown for a run, from its run log
list_campaignsList campaigns visible to the dashboard user (read only)
get_findingsVulnerabilities and severity statistics for a campaign (read only)

Only run assessments against systems you own or are explicitly authorized in writing to test. Findings can include false positives and must be reviewed by a qualified human.

Configuration

VariableDescription
DARKMOON_BASE_URLBase URL of your Darkmoon Pro Dashboard API (required)
DARKMOON_USERNAME, DARKMOON_PASSWORDDashboard credentials; a JWT is requested on each call and never cached
DARKMOON_TOKENAlternative to username/password: a pre-issued JWT
DARKMOON_TIMEOUT_MSOptional per-request timeout, default 60000

Client configuration

Claude Desktop (claude_desktop_config.json), Continue and LibreChat use the same mcpServers shape:

json
{
  "mcpServers": {
    "darkmoon": {
      "command": "npx",
      "args": ["-y", "@darkmoon_ai/mcp-server"],
      "env": {
        "DARKMOON_BASE_URL": "https://darkmoon.example.internal",
        "DARKMOON_USERNAME": "your-dashboard-user",
        "DARKMOON_PASSWORD": "your-dashboard-password"
      }
    }
  }
}

Goose (~/.config/goose/config.yaml):

yaml
extensions:
  darkmoon:
    type: stdio
    enabled: true
    name: darkmoon
    cmd: npx
    args: ["-y", "@darkmoon_ai/mcp-server"]
    envs:
      DARKMOON_BASE_URL: https://darkmoon.example.internal
      DARKMOON_USERNAME: your-dashboard-user
      DARKMOON_PASSWORD: your-dashboard-password

Continue (.continue/mcpServers/darkmoon.yaml):

yaml
name: Darkmoon
version: 0.1.0
schema: v1
mcpServers:
  - name: darkmoon
    command: npx
    args: ["-y", "@darkmoon_ai/mcp-server"]
    env:
      DARKMOON_BASE_URL: https://darkmoon.example.internal
      DARKMOON_USERNAME: your-dashboard-user
      DARKMOON_PASSWORD: your-dashboard-password

Develop

bash
npm install
npm run build
npm test      # mocked Dashboard API, in-memory MCP client and a real stdio process

License

GPL-3.0-only, same as Darkmoon.

Similar modules

Browse Penetration testing