First, create the virtual environment using uv and install both MCP servers into it. The default path used here is ~/.local/envs/.venv_mcp — adjust this path if you want your env somewhere else.

# Create the virtual environment
uv venv ~/.local/envs/.venv_mcp

# Install both MCP servers into it
uv pip install --python ~/.local/envs/.venv_mcp root-mcp cerngitlab-mcp

After this, both servers are available as executables inside ~/.local/envs/.venv_mcp/bin/.

Qwen Code reads MCP servers from mcpServers in settings.json. You have two scopes: qwenlm.github

• Project scope (only current project): .qwen/settings.json in your project root
• User scope (all projects): ~/.qwen/settings.json

Create or edit your chosen settings.json and add the following:

{
  "mcpServers": {
    "root-mcp": {
      "command": "/home/YOUR_USERNAME/.local/envs/.venv_mcp/bin/root-mcp",
      "args": ["--data-path", "/path/to/your/root/files"],
      "timeout": 30000
    },
    "cerngitlab": {
      "command": "/home/YOUR_USERNAME/.local/envs/.venv_mcp/bin/cerngitlab-mcp",
      "args": []
    }
  }
}

Adjustments to make:

• Replace /home/YOUR_USERNAME/ with your actual home directory (run echo $HOME to find it). On Linux this is typically /home/yourname, on macOS /Users/yourname.

• Replace /path/to/your/root/files with the directory containing your .root files.

• If you placed the env in a different location than ~/.local/envs/.venv_mcp, update the paths accordingly.

• If you want the servers available globally across all projects, write to ~/.qwen/settings.json instead.

By default cerngitlab-mcp only accesses public repositories. To unlock private repos and additional tools like search_code and search_issues, add your CERN GitLab token:

"cerngitlab": {
  "command": "/home/YOUR_USERNAME/.local/envs/.venv_mcp/bin/cerngitlab-mcp",
  "args": [],
  "env": {
    "CERNGITLAB_TOKEN": "glpat-xxxxxxxxxxxx"
  }
}

Generate a token at https://gitlab.cern.ch/-/user_settings/personal_access_tokens with the read_api scope. You can also reference an existing shell variable instead of hardcoding: "CERNGITLAB_TOKEN": "$CERNGITLAB_TOKEN".

Restart Qwen Code in your project directory, then run:

qwen mcp list

Both root-mcp and cerngitlab should appear as connected. If a server shows "Disconnected", double-check that the command path is the exact absolute path to the binary inside your venv, you can verify with:

ls ~/.local/envs/.venv_mcp/bin/