Skip to main content

MCP

@baleybots/mcp is a thin MCP (Model Context Protocol) client for Baleybots processors. It wraps @ai-sdk/mcp and adapts remote tools into the shape Baleybot expects: { name, description, inputSchema, execute }.

Tools returned by mcpTools() use execute — the deprecated function field was removed in core alpha.171.

Install​

bun add @baleybots/mcp @baleybots/core ai zod

Dependencies: @ai-sdk/mcp ^2.0.9
Peer dependencies: @baleybots/core >=0.0.1-alpha.100, ai ^7.0.0, zod ^4.3.6

Exports​

ExportPurpose
MCPClientConnect to one MCP server (stdio, HTTP/SSE)
MCPManagerManage multiple servers, aggregate tools, idle disconnect
DockerStdioTransportRun MCP servers in Docker via stdio
mcpToolsAdapt a connected client to Baleybot-compatible tools

Connect and attach tools​

import { Baleybot } from '@baleybots/core';
import { MCPClient, mcpTools } from '@baleybots/mcp';

const client = new MCPClient({
  name: 'filesystem',
  transport: {
    type: 'stdio',
    command: 'npx',
    args: ['-y', '@modelcontextprotocol/server-filesystem', '/tmp/workspace'],
  },
});

await client.connect();
const tools = await mcpTools(client);

const bot = Baleybot.create({
  name: 'file-processor',
  goal: 'Help with files in the MCP workspace.',
  tools,
});

await bot.process('List the top-level files and summarize each one.');
await client.disconnect();

Prefix or filter tools when adapting:

const tools = await mcpTools(client, {
  prefix: 'fs_',
  filter: (t) => t.name.includes('read'),
});

Multiple servers​

MCPManager connects several servers, caches tools, and auto-disconnects idle clients:

import { MCPManager } from '@baleybots/mcp';

const manager = new MCPManager();

await manager.addServer({
  name: 'github',
  transport: {
    type: 'stdio',
    command: 'npx',
    args: ['-y', '@modelcontextprotocol/server-github'],
    env: { GITHUB_TOKEN: process.env.GITHUB_TOKEN! },
  },
  toolPrefix: 'github_',
});

const tools = await manager.getTools();

const bot = Baleybot.create({
  name: 'dev',
  goal: 'Help with repos and local files.',
  tools,
});

await manager.disconnectAll();

Resources and prompts​

MCPClient also exposes AI SDK client methods: listTools, callTool, listResources, readResource, listPrompts, getPrompt. Use these when you need MCP primitives outside the tool loop.