@beautiful-ccg/mcp-server

MCP server that exposes bccg capabilities as tools via the Model Context Protocol.

Usage

import { createServer, startServer } from "@beautiful-ccg/mcp-server";
import { Orchestrator, Registry } from "@beautiful-ccg/core";

const orchestrator = new Orchestrator(registry);

// Option 1: Start stdio server (blocks)
await startServer(orchestrator);

// Option 2: Create server for custom transport
const server = createServer(orchestrator);
await server.connect(myTransport);

MCP Tools

bccg_run

Run a prompt through multiple AI CLIs with auto-routing.

| Parameter | Type | Required | Description | |---|---|---|---| | prompt | string | yes | The prompt to run | | strategy | enum | no | cheap-first, quality-first, balanced, parallel | | adapter | string | no | Specific adapter name | | model | string | no | Model for intra-CLI routing |

bccg_pipeline

Execute a multi-step CCG pipeline.

| Parameter | Type | Required | Description | |---|---|---|---| | steps | string | yes | Pipeline DSL (e.g. gemini:summarize -> claude:review) | | prompt | string | no | Base prompt for the pipeline |

bccg_status

Check available CLI adapters and their status. No parameters. Returns JSON.

Recursion Guard

Both bccg_run and bccg_pipeline are blocked when BCCG_DEPTH >= 1, returning an error. This prevents infinite loops when a host CLI invokes bccg which would spawn the same CLI.

Transport

Uses StdioServerTransport from @modelcontextprotocol/sdk. Handles SIGINT and SIGTERM for clean shutdown.

Exports

• createServer(orchestrator): McpServer — create an MCP server instance
• startServer(orchestrator): Promise<void> — start stdio server (blocks until signal)