OpenMetadata

You can connect your MCP clients to the OpenMetadata data platform through a dedicated MCP server that exposes standardized endpoints for metadata entities and assets. This server acts as a bridge, letting you manage tables, dashboards, topics, pipelines, and more via a consistent MCP interface while OpenMetadata handles the underlying data storage and API calls.

How to use

To work with the OpenMetadata MCP Server, run it as a local service or start it via a client command, then point your MCP-enabled client at the server. You can operate core data entities like tables, databases, and database schemas as well as assets such as dashboards, charts, pipelines, topics, metrics, containers, reports, and ML models. You can also manage users, teams, classifications, glossaries, tags, policies, and domains. The server supports both HTTP-based remote interaction and local stdio-based execution, depending on how you choose to run and configure it.

How to install

Prerequisites: you need Python installed on your system to run the MCP server locally. You may also install client tooling as needed. Follow the concrete steps below to install and start the server.

# Install Python if it is not already installed
# Example for Debian/Ubuntu
sudo apt-get update
sudo apt-get install -y python3 python3-venv python3-pip

# (Optional) Set up a virtual environment
python3 -m venv venv
source venv/bin/activate

# Install dependencies if any package manifest exists (adjust as needed)
pip install -r requirements.txt  # if a requirements file is provided

# Run the server directly from source
python src/server.py

Configuration and usage notes

You can run the MCP server in different ways depending on how you want to integrate it with your workflow. The server can be invoked via a local Python command, or connected through MCP client configurations that use either a remote HTTP approach or a local stdio launcher. When using local stdio, you specify the command and arguments to start the server so the client can communicate with it over standard input/output.

{
  "mcpServers": {
    "mcp-server-openmetadata": {
      "command": "uvx",
      "args": ["mcp-server-openmetadata"],
      "env": {
        "OPENMETADATA_HOST": "https://your-openmetadata-host",
        "OPENMETADATA_JWT_TOKEN": "your-jwt-token"
      }
    }
  }
}

Additional MCP client configurations

If you prefer a different launcher, you can use an alternative configuration that points to the same MCP server. The following shows another launcher example with a directory-based run command.

{
  "mcpServers": {
    "mcp-server-openmetadata": {
      "command": "uv",
      "args": [
        "--directory",
        "/path/to/mcp-server-openmetadata",
        "run",
        "mcp-server-openmetadata"
      ],
      "env": {
        "OPENMETADATA_HOST": "https://your-openmetadata-host",
        "OPENMETADATA_JWT_TOKEN": "your-jwt-token"
      }
    }
  }
}

Manual start for local development

You can start the server directly with Python from the source if you prefer a straightforward development run.

python src/server.py

Options:

• --port: Port to listen on for SSE (default: 8000)
• --transport: Transport type (stdio/sse, default: stdio)

## Security and credentials

When connecting from MCP clients, provide authentication details as shown in the configuration examples. Use a secure host endpoint and keep tokens or credentials private. If you are using token-based authentication, set OPENMETADATA\_HOST and OPENMETADATA\_JWT\_TOKEN; for basic authentication, set OPENMETADATA\_HOST along with OPENMETADATA\_USERNAME and OPENMETADATA\_PASSWORD.

## Troubleshooting and tips

If the server fails to start, check that Python and any required dependencies are installed, and verify that the host URL is reachable. Ensure that the environment variables are set correctly in your MCP launcher configuration. If you are using a directory-based launcher, confirm that the path to the server directory is correct and that you are running the intended entry point.