pulseflow-mcp

A local Model Context Protocol server for PulseFlow — a problems-first Flutter performance lab. It lets a coding agent (Cursor, Claude Code, Command Code) read a live Flutter session and find, fix, and verify performance problems without copy-pasting reports.

The server connects to your running PulseFlow bridge and exposes the same ranked problems, health score, and agent report the dashboard shows.

Prerequisites

• Node 18+
• The PulseFlow bridge running (from the PulseFlow repo: make run, or make bridge).
• Your Flutter app running (debug for rebuilds/source, profile for CPU).

Use it

No install needed — point your MCP client at npx:

Cursor — .cursor/mcp.json (or ~/.cursor/mcp.json):

{
  "mcpServers": {
    "pulseflow": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "pulseflow-mcp"]
    }
  }
}

Command Code — cmd mcp add pulseflow -- npx -y pulseflow-mcp.

Then restart the client and, in the agent, run pulseflow_connect (or pass {"mock": true} for a device-free demo) and ask:

What are the top performance problems right now, and what should I fix first?

Tools

• Session: pulseflow_status, pulseflow_discover, pulseflow_connect, pulseflow_disconnect
• Findings: pulseflow_report, pulseflow_problems, pulseflow_widgets, pulseflow_frames, pulseflow_focus_snapshot, pulseflow_cpu, pulseflow_memory, pulseflow_network, pulseflow_device, pulseflow_errors, pulseflow_logs
• Baselines: pulseflow_snapshot (save "before"), pulseflow_compare (diff "before" → "after")
• Controls: pulseflow_debug_option, pulseflow_hot_widgets, pulseflow_reset_telemetry

Configuration

| Env var | Default | Meaning | | --- | --- | --- | | PULSEFLOW_BRIDGE_URL | ws://127.0.0.1:3847/bridge | Bridge WebSocket to connect to | | PULSEFLOW_VM_URL | – | Optional VM Service URL to auto-connect at startup |

License

MIT