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 use OpenAI client with Exa search capabilities.

npm install @smithery/mcp-exa

The following code sets up OpenAI and connects to an Exa MCP server. In this case, we're running the server locally within the same process, so it's just a simple passthrough.

import { MultiClient } from "@smithery/sdk"
import { OpenAIChatAdapter } from "@smithery/sdk/integrations/llm/openai"
import * as exa from "@smithery/mcp-exa"
import { OpenAI } from "openai"
import { createTransport } from "@smithery/sdk/registry"

const openai = new OpenAI()
const exaServer = exa.createServer({
  apiKey: process.env.EXA_API_KEY,
})

const sequentialThinking = await createTransport(
  "@modelcontextprotocol/server-sequential-thinking",
)
const client = new MultiClient()
await client.connectAll({
  exa: exaServer,
  sequentialThinking: sequentialThinking,
})

Now you can make your LLM aware of the available tools from Exa.

// Create an adapter
const adapter = new OpenAIChatAdapter(client)
const response = await openai.chat.completions.create({
  model: "gpt-4o-mini",
  messages: [{ role: "user", content: "In 2024, did OpenAI release GPT-5?" }],
  // Pass the tools to OpenAI call
  tools: await adapter.listTools(),
})
// Obtain the tool outputs as new messages
const toolMessages = await adapter.callTool(response)

Using this, you can easily enable your LLM to call tools and obtain the results.

However, it's often the case where your LLM needs to call a tool, see its response, and continue processing output of the tool in order to give you a final response.

In this case, you have to loop your LLM call and update your messages until there are no more toolMessages to continue.

Example:

let messages = [
  {
    role: "user",
    content:
      "Deduce Obama's age in number of days. It's November 28, 2024 today. Search to ensure correctness.",
  },
]
const adapter = new OpenAIChatAdapter(client)

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.

Troubleshooting

Error: ReferenceError: EventSource is not defined

This event means you're trying to use EventSource API (which is typically used in the browser) from Node. You'll have to install the following to use it:

npm install eventsource
npm install -D @types/eventsource

Patch the global EventSource object:

import EventSource from "eventsource"
global.EventSource = EventSource as any