HexagonML ModelManager MCP Server

HexagonML ModelManager MCP Server

This is a HexagonML MCP server that provides a Model context protocol interface for HexagonML ModelManager tools.

Local Development

Prerequisites

• Python: python3

• Virtualenv: recommended (project uses .venv in examples)

Environment variables

The server reads configuration from environment variables (and will also load a local .env file automatically).

• SECRET_KEY: secret key for ModelManager API auth

• MM_API_BASE_URL: base URL for ModelManager API (example: http://localhost:8000)

• OUTPUT_DIR: directory where generated HTML outputs are written

• HOST (optional): defaults to 0.0.0.0

• PORT (optional): defaults to 9000

Example .env:

SECRET_KEY=your-secret-key
MM_API_BASE_URL=http://localhost:8000
OUTPUT_DIR=./output
HOST=0.0.0.0
PORT=9000

Run the server

From the repo root:

python3 server/mm_mcp_server.py

To run with the FastMCP inspector (dev mode):

fastmcp dev server/mm_mcp_server.py

Troubleshooting

• Port in use (FastMCP inspector): If you see Proxy Server PORT IS IN USE (commonly 6277), stop the previous inspector process and retry.

• Missing env vars: The server will exit with a message listing missing required variables.

Configuration For mcp integration on host (g: windsurf, vscode, claude desktop)

Local Configuration

{
  "mcpServers": {
    "hex-mm-mcp": {
      "command": "hex-mm-mcp/.venv/bin/mcp",
      "args": ["run", "hex-mm-mcp/server/mm_mcp_server.py"]
    }
  }
}

Docker Configuration

For Dev (Using Local URL)

• Run the ModelManager server in --host 0.0.0.0 --port 8000

  • cmd python manage.py runserver 0.0.0.0:8000

• Get the hostname of your system using hostname -I command

  • eg: 192.168.10.75 172.17.0.1 2400:1a00:4b26:2af0:8f53:ede1:ec3a:c59b 2400:1a00:4b26:2af0:9139:c926:2fb5:6008

  • use first ip address from the list eg: 192.168.10.75

• Replace your-api-base-url with http://<hostIP>:8000

{
  "mcpServers": {
    "hex-mm-mcp-docker": {
      "command": "docker",
      "args": [
        "run",
        "--rm",
        "-i",
        "--network=host",
        "-e",
        "SECRET_KEY",
        "-e",
        "MM_API_BASE_URL",
        "-v",
        "OUTPUT_DIR",
        "<image-name>:<tag>"
      ],
      "env": {
        "SECRET_KEY": "your-secret-key",
        "MM_API_BASE_URL": "your-api-base-url",
        "OUTPUT_DIR": "your-output-dir"
      }
    }
  }
}

Docker Commands

Build Image
docker build --platform=linux/amd64 -t modelmanagerdev/mcp:version_id .

Run Container
docker run --platform=linux/amd64  -d --name mm-mcp -p 9000:9000 --env-file .env modelmanagerdev/mcp:v6

Model Insights Tools

The server exposes Model Insights endpoints via MCP tools.

Create Insight

• Tool name: create_insight

• Input: data (dict)

• Backend: POST /api/mmanager-modelinsights/create_insight/

Get Insights

• Tool name: get_insight

• Input: usecase_id (str)

• Backend: GET /api/mmanager-modelinsights/get_insights/?usecase_id=...

• Response shape:

  • If the backend returns a dict, the tool returns that dict.

  • If the backend returns a list, the tool returns { "data": [...] }.