MCP Protocol

How it works

1. Add an MCP server — go to Settings → Integrations → MCP Servers in your dashboard and register your server’s URL (or POST /api/mcp via the API).
2. Agent discovers tools — when your agent runs, GWC exposes the registered MCP tools automatically.
3. Agent calls tools — the agent decides which tool to use, sends a request, and gets structured results back.
4. Where tools run — Taichu runs MCP tools locally on your machine. Cloud-registered MCP servers are invoked by GWC’s API on your behalf — tool inputs and outputs pass through GWC infrastructure.

Connect your own MCP servers

Register public MCP endpoints from your dashboard (Settings → Integrations → MCP Servers) or via POST /api/mcp, and add Taichu-local servers with the gwc.mcpServers setting.

Custom MCP server

Add your own MCP-compatible tool server in the Taichu settings for GWC for VS Code:

// Example: custom MCP server config for Taichu in GWC for VS Code
{
  "mcpServers": {
    "my-tool": {
      "command": "node",
      "args": ["./my-mcp-server.js"],
      "env": {
        "API_KEY": "${env:MY_API_KEY}"
      }
    }
  }
}

Creating tasks via SDK

Create tasks programmatically — when your agent runs, it discovers the MCP tools registered to your workspace automatically:

// Create a task via the SDK
const gwc = new GWC({ apiKey: 'your-key' });

await gwc.task.create({
  name: 'Analyze database',
  agent: 'data-analyst',
});