Skip to main content

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.

Build the MCP server​

cargo build -p pytxo-mcp --release

Configure Cursor​

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

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

Adjust paths for your OS. PYTXO_REPO should point at the git root where pytxo.toml lives.

What MCP enables​

  • Trigger pytxo run and preflight from the agent context
  • Query run status and agent topology
  • Tail structured logs without opening a separate terminal grid

Pytxo remains local-first: the MCP process talks to the Rust orchestrator on your machine. No cloud workspace required.

Troubleshooting​

  1. Run pytxo doctor in the target repo before MCP-driven runs.
  2. Ensure the MCP binary matches the CLI version you built from the same commit.
  3. Check .pytxo/data/ is writable.

See CLI reference for flags your MCP tools may wrap.