MCP API - NeuralCleave

The NeuralCleave MCP server implements Model Context Protocol over stdio (JSON-RPC 2.0). Once spawned, external MCP clients (Claude Desktop, Cursor, VS Code extensions) can call NeuralCleave’s registered tools directly.

POST /mcp/spawn

Start the MCP stdio server subprocess.

curl -X POST http://localhost:7432/api/v1/mcp/spawn

{
  "pid": 12345,
  "running": true,
  "already_running": false
}

Calling this endpoint when the server is already running returns "already_running": true without starting a second process.

GET /mcp/status

Check whether the MCP server is running.

curl http://localhost:7432/api/v1/mcp/status

{
  "running": true,
  "pid": 12345
}

DELETE /mcp/server

Stop the MCP stdio server.

curl -X DELETE http://localhost:7432/api/v1/mcp/server

{ "killed": true }

MCP protocol

The stdio server implements MCP protocol version 2024-11-05:

Method	Description
`initialize`	Handshake; returns capabilities and server info
`tools/list`	List all registered NeuralCleave tools as MCP tool descriptors
`tools/call`	Call a tool by name with arguments
`ping`	Liveness check

Notifications (requests without an id) are silently ignored per the MCP spec.

Claude Desktop integration

Add to ~/.config/claude/config.json:

{
  "mcpServers": {
    "neuralcleave": {
      "command": "python",
      "args": ["-m", "neuralcleave.mcp.server"]
    }
  }
}

Restart Claude Desktop. NeuralCleave tools appear in the tool panel.

Cursor integration

Add to .cursor/mcp.json in your project root:

{
  "mcpServers": {
    "neuralcleave": {
      "command": "python",
      "args": ["-m", "neuralcleave.mcp.server"]
    }
  }
}

​POST /mcp/spawn

​GET /mcp/status

​DELETE /mcp/server

​MCP protocol

​Claude Desktop integration

​Cursor integration