Smithery Typescript Framework

Smithery is a Typescript framework with utilities to make it easier to connect language models (LLMs) to Model Context Protocols (MCPs), allowing you to build agents that use resources and tools without being overwhelmed by JSON schemas.

⚠️ This repository is work in progress and in alpha. Not recommended for production use yet. ⚠️

Key Features

• Connect to multiple MCPs with a single client
• Adapters to transform MCP responses for OpenAI and Anthropic clients
• Supports chaining tool calls until LLM completes

To find our registry of MCPs, visit https://smithery.ai/.

Quickstart

Installation

npm install @smithery/sdk

Usage

In this example, we'll connect to Exa search capabilities using either OpenAI or Anthropic.

npm install @smithery/sdk @modelcontextprotocol/sdk

The following code sets up the client and connects to an Exa MCP server:

import { MultiClient } from "@smithery/sdk"
import { OpenAIChatAdapter } from "@smithery/sdk/integrations/llm/openai"
import { AnthropicChatAdapter } from "@smithery/sdk/integrations/llm/anthropic"
import { createTransport } from "@smithery/sdk/transport.js"
import { OpenAI } from "openai"
import Anthropic from "@anthropic-ai/sdk"

// Create a new connection
const exaTransport = createTransport(
  // Replace with your deployed MCP server URL
  "https://your-mcp-server.example.com"
)

// Initialize a multi-client connection
const client = new MultiClient()
await client.connectAll({
  exa: exaTransport,
  // You can add more connections here...
})

// Configure and authenticate
await client.clients.exa.request({
  method: "config",
  params: {
    config: {
      apiKey: process.env.EXA_API_KEY,
    },
  },
})

Now you can use either OpenAI or Anthropic to interact with the tools:

// Using OpenAI
const openai = new OpenAI()
const openaiAdapter = new OpenAIChatAdapter(client)
const openaiResponse = await openai.chat.completions.create({
  model: "gpt-4o-mini",
  messages: [{ role: "user", content: "What AI events are happening in Singapore?" }],
  tools: await openaiAdapter.listTools(),
})
const openaiToolMessages = await openaiAdapter.callTool(openaiResponse)

For more complex interactions where the LLM needs to process tool outputs and potentially make additional calls, you'll need to implement a conversation loop. Here's an example:

let messages = [
  {
    role: "user",
    content: "What are some AI events happening in Singapore and how many days until the next one?",
  },
]
const adapter = new OpenAIChatAdapter(client)
let isDone = false

while (!isDone) {
  const response = await openai.chat.completions.create({
    model: "gpt-4o-mini",
    messages,
    tools: await adapter.listTools(),
  })
  
  // Handle tool calls
  const toolMessages = await adapter.callTool(response)

  // Append new messages
  messages.push(response.choices[0].message)
  messages.push(...toolMessages)
  isDone = toolMessages.length === 0
}

See a full example in the examples directory.