AISDK MCP Bridge

A bridge package that enables seamless integration between the Model Context Protocol (MCP) and AI SDK, allowing for efficient communication and tool execution between MCP servers and AI models.

Features

- Seamless integration between MCP servers and AI SDK
- Support for various MCP server types (Node.js, Python, UVX)
- Multi-server support with independent configuration
- Flexible configuration through mcp.config.json
- TypeScript support with full type definitions
- Robust error handling and logging
- Easy-to-use API for tool execution

Installation

npm install aisdk-mcp-bridge

Quick Start

1. Create an mcp.config.json file in your project root:

{
  "mcpServers": {
    "twitter-mcp": {
      "command": "npx",
      "args": ["-y", "@enescinar/twitter-mcp"],
      "env": {
        "API_KEY": "your-twitter-api-key",
        "API_SECRET_KEY": "your-twitter-api-secret",
        "ACCESS_TOKEN": "your-twitter-access-token",
        "ACCESS_TOKEN_SECRET": "your-twitter-access-token-secret"
      }
    },
    "firecrawl": {
      "command": "npx",
      "args": ["-y", "mcp-server-firecrawl"],
      "env": {
        "FIRE_CRAWL_API_KEY": "your-firecrawl-api-key",
        "FIRE_CRAWL_API_URL": "https://api.firecrawl.com"
      }
    }
  }
}

2. Import and use the bridge in your code:

import { generateText } from 'ai';
import { google } from '@ai-sdk/google';
import { getMcpTools, cleanupMcp, initializeMcp } from 'aisdk-mcp-bridge';
import dotenv from 'dotenv';
dotenv.config();
async function main() {

  try {

    // Initialize MCP

    await initializeMcp({ debug: true });
// Get tools from all servers

    const allTools = await getMcpTools({ debug: true });
// Or get tools from a specific server

    const twitterTools = await getMcpTools({

      debug: true,

      serverName: 'twitter-mcp',

    });
// Use tools with AI SDK

    const result = await generateText({

      model: google('gemini-1.5-pro'),

      messages: [

        {

          role: 'system',

          content:

            'You are an AI assistant that uses various tools to help users.',

        },

        {

          role: 'user',

          content: 'Your task description here',

        },

      ],

      tools: twitterTools, // or allTools for all available tools

    });
console.log('Result:', result.text);

  } finally {

    // Clean up resources

    await cleanupMcp();

  }

}
main().catch(error => {

  console.error('Error:', error);

  process.exit(1);

});

Configuration

The mcp.config.json file supports multiple servers and communication modes. Each server can be configured independently.

Server Configuration Examples:

Twitter MCP Server

{
  "mcpServers": {
    "twitter-mcp": {
      "command": "npx",
      "args": ["-y", "@enescinar/twitter-mcp"],
      "env": {
        "API_KEY": "your-twitter-api-key",
        "API_SECRET_KEY": "your-twitter-api-secret",
        "ACCESS_TOKEN": "your-twitter-access-token",
        "ACCESS_TOKEN_SECRET": "your-twitter-access-token-secret"
      }
    }
  }
}

Firecrawl Server

{
  "mcpServers": {
    "firecrawl": {
      "command": "npx",
      "args": ["-y", "mcp-server-firecrawl"],
      "env": {
        "FIRE_CRAWL_API_KEY": "your-firecrawl-api-key",
        "FIRE_CRAWL_API_URL": "https://api.firecrawl.com"
      }
    }
  }
}

SSE Server

{
  "mcpServers": {
    "sse-server": {
      "command": "node",
      "args": ["./server.js"],
      "mode": "sse",
      "sseOptions": {
        "endpoint": "http://localhost:3000/events",
        "headers": {},
        "reconnectTimeout": 5000
      }
    }
  }
}

Server Modes

The bridge supports different communication modes:

1. stdio Mode (Default)

- Direct communication through standard input/output
- Best for simple integrations and local development
- Low latency and minimal setup required

2. SSE Mode (Server-Sent Events)
- Real-time, one-way communication from server to client
- Ideal for streaming updates and long-running operations
- Built-in reconnection handling

API Reference

Core Functions

initializeMcp(options?: InitOptions): Promise<void>

Initialize the MCP service with the provided options.

interface InitOptions {
  configPath?: string; // Path to mcp.config.json
  debug?: boolean; // Enable debug logging
}

getMcpTools(options?: ToolOptions): Promise<ToolSet>

Get AI SDK-compatible tools from MCP servers.

interface ToolOptions {
  debug?: boolean; // Enable debug logging
  serverName?: string; // Optional server name to get tools from a specific server
}

executeMcpFunction(serverName: string, functionName: string, args: Record<string, unknown>): Promise<MCPToolResult>

Execute a specific function on an MCP server directly.

// Example
const result = await executeMcpFunction('twitter-mcp', 'postTweet', {
  text: 'Hello from MCP!',
});

Core Types

MCPConfig (alias for MCPServersConfig)

Configuration type for MCP servers.

interface MCPConfig {
  mcpServers: {
    [key: string]: ServerConfig;
  };
}

ServerConfig

Configuration for individual MCP servers.

interface ServerConfig {
  command: string;
  args?: string[];
  env?: Record<string, string>;
  mode?: 'stdio' | 'sse';
  sseOptions?: {
    endpoint: string;
    headers?: Record<string, string>;
    reconnectTimeout?: number;
  };
}

MCPToolResult

Result type for MCP tool executions.

interface MCPToolResult {
  success: boolean;
  data?: unknown;
  error?: string;
}

cleanupMcp(): Promise<void>

Clean up MCP resources and close all server connections.

Error Handling

The bridge includes comprehensive error handling for:

- Server initialization failures
- Communication errors
- Tool execution failures
- Configuration issues
- Server connection issues

Logging

The bridge provides detailed logging through:

- mcp-tools.log: Server-side tool execution logs
- Console output for debugging and errors

Debug Logging

You can enable detailed debug logging by setting the DEBUG environment variable:

```bash