MCP from Cursor

Pytxo exposes orchestration through a local MCP server so Cursor (or any MCP-capable IDE) can start runs, inspect status, and tail logs without replacing your editor.

Install

1. Install the CLI (npm i -g pytxo or a release binary).
2. Confirm the MCP binary is available. The npm package ships pytxo-mcp next to pytxo. Find it with:

# npm global (example paths)
which pytxo-mcp        # macOS / Linux
where.exe pytxo-mcp    # Windows

If you only have the CLI zip from GitHub Releases, download the matching platform archive that includes pytxo-mcp, or build from source under Developers.

Configure Cursor

Add a local MCP server entry (.cursor/mcp.json or your global MCP config):

{
  "mcpServers": {
    "pytxo": {
      "command": "pytxo-mcp",
      "args": [],
      "env": {
        "PYTXO_REPO": "/absolute/path/to/your/project"
      }
    }
  }
}

On Windows, set command to the full path of pytxo-mcp.exe if it is not on PATH. PYTXO_REPO must be the git root where pytxo.toml lives.

What MCP enables

• Trigger runs and dry-runs from the agent context
• Query status and live agents
• Tail structured logs
• Route stdin to live agents
• Start modular project and fleet runs

Full tool list: MCP tools.

Troubleshooting

1. Run pytxo doctor in the target repo before MCP-driven runs.
2. Keep pytxo and pytxo-mcp on the same release.
3. Ensure .pytxo/data/ is writable.
4. See Troubleshooting for trust blocks and path mismatches.