DefectDojo Community

DefectDojo Community MCP Server

[CI] [PyPI] [Python] [License] [OpenSSF Scorecard]

A read-only Model Context Protocol server for DefectDojo Community Edition — so an AI assistant can read your vulnerability findings, group the same vulnerability across every product, and draft issues for whatever tracker you use.

Unofficial. This is an independent open-source project. It is not affiliated with, endorsed by, or supported by DefectDojo Inc. or OWASP. "DefectDojo" is used only to describe what this server connects to.

Do you actually need this?

If you run DefectDojo Pro, you probably don't. Pro has a built-in MCP server — use that instead; it is maintained by the vendor and more capable.

This project exists because Community Edition has no MCP server at all. It talks to the ordinary REST API v2 with a normal API token, so it needs nothing enabled server-side and no particular DefectDojo version.

Not to be confused with the unrelated defectdojo-mcp package on PyPI, which is a different third-party project. This one is defectdojo-community-mcp.

Tools

"Actionable" means active and not false-positive, duplicate, risk-accepted, out-of-scope or mitigated — the set actually worth someone's time.

group_critical_by_vuln is the reason this exists. A monorepo-shaped scanner setup reports the same upstream CVE once per product; this collapses them into one group per vulnerability listing every affected product, which maps cleanly onto one tracker issue.

Install

Requires Python 3.10+.

pip install defectdojo-community-mcp

Or as a container:

docker pull ghcr.io/systechgd/defectdojo-community-mcp:latest

Listed on the MCP Registry as io.github.systechgd/defectdojo-community-mcp, so MCP clients that read the registry can install it by name.

Or from source:

git clone https://github.com/systechgd/defectdojo-community-mcp.gitcd defectdojo-community-mcppip install -e .

Configure

Two environment variables are required.

Get a token from DefectDojo → your user → API v2 Key.

Create a dedicated read-only user for it. The token inherits that user's permissions, and nothing in this server needs write access. If the token leaks, a read-only one cannot be used to alter your findings.

Check it works before wiring up a client:

defectdojo-community-mcp --check# ok: https://defectdojo.example.com reachable, 23 product(s) visible

Connect an MCP client

Claude Code

claude mcp add defectdojo \  --env DEFECTDOJO_URL=https://defectdojo.example.com \  --env DEFECTDOJO_API_KEY=your-token \  -- defectdojo-community-mcp

Claude Desktop / generic JSON config

{  "mcpServers": {    "defectdojo": {      "command": "defectdojo-community-mcp",      "env": {        "DEFECTDOJO_URL": "https://defectdojo.example.com",        "DEFECTDOJO_API_KEY": "your-token"      }    }  }}

If the console script is not on your PATH, use "command": "python", "args": ["-m", "defectdojo_community_mcp"].

Docker

The server speaks MCP over stdio, so run it attached (-i), not detached:

{  "mcpServers": {    "defectdojo": {      "command": "docker",      "args": [        "run", "--rm", "-i",        "-e", "DEFECTDOJO_URL",        "-e", "DEFECTDOJO_API_KEY",        "ghcr.io/systechgd/defectdojo-community-mcp:latest"      ]    }  }}

Example

Group every critical finding by vulnerability and show me the ones affecting more than three services.

The assistant calls group_critical_by_vuln and gets back, per vulnerability, the CVE, title, affected product list and a direct DefectDojo URL per finding — enough to write one issue covering all of them, instead of twelve near-identical ones.

Safety model

Read-only is a design invariant, not a default.

• No tool issues a POST, PUT, PATCH or DELETE.
• The HTTP client class exposes no method capable of doing so, so a mutating tool cannot be added by accident — only by deliberately changing that class.
• A test asserts that every request made by every tool is a GET.

Adding write support would be a breaking change to this guarantee. If it ever happens it will be a separate server, so operators can keep granting read without granting write.

Findings data is attacker-influenced — scanner output can contain text from dependency metadata, file paths and commit messages. Treat tool results as data, never as instructions. See SECURITY.md for the full threat model and how to report a vulnerability.

Development

git clone https://github.com/systechgd/defectdojo-community-mcp.gitcd defectdojo-community-mcppython -m venv .venv && . .venv/bin/activate   # Windows: .venv\Scripts\activatepip install -e ".[dev]"
pytest          # 57 tests, no network or live instance neededruff check .

The test suite runs entirely against httpx.MockTransport. CI additionally runs the suite on Python 3.10–3.13 and against both MCP SDK generations (mcp<2, which has FastMCP, and mcp>=2, which renamed it to MCPServer).

See CONTRIBUTING.md and CODE_OF_CONDUCT.md.

License

Apache-2.0. See NOTICE.