mcp-use - Fullstack MCP Framework for TypeScript and Python

TL;DR

TL;DR: mcp-use is an open-source SDK (10k GitHub stars, MIT license) for building MCP servers and cross-client apps in TypeScript and Python, with an optional Manufact Cloud deployment platform.

Source and Accuracy Notes

⚠️ This section is MANDATORY. All links must be verified from actual source, not guessed.

• Project page: manufact.com ← verified product homepage
• Source repository: github.com/mcp-use/mcp-use ← verified README
• License: MIT ← verified in repo
• HN launch thread: news.ycombinator.com/item?id=48110929
• Source last checked: 2026-06-23 (commit verified via GitHub API, 10,134 stars)

What Is mcp-use?

mcp-use is the fullstack MCP framework built by Manufact for developing MCP servers (for AI agents) and MCP apps (for ChatGPT and Claude). The SDK supports both TypeScript and Python, letting you define tools with Zod schemas and deploy behind the Manufact MCP Cloud or any HTTP endpoint.

From the README:

mcp-use is the fullstack MCP framework to build MCP Apps for ChatGPT / Claude & MCP Servers for AI Agents.

Key capabilities:

• TypeScript + Python SDKs with first-class Zod schema validation
• MCP Inspector — a browser-based tool for testing and debugging servers locally (http://localhost:3000/inspector)
• MCP Apps — interactive React widgets that run across ChatGPT, Claude, and other MCP clients
• Branch deployments on Manufact Cloud with metrics, logs, and observability
• Skills integration — install as a skill for Claude Code, Codex, or Cursor

Setup Workflow

Step 1: Create a new MCP server

The TypeScript quickstart uses create-mcp-use-app:

npx create-mcp-use-app@latest

Or initialize manually:

npm install mcp-use

Step 2: Define a tool with Zod schema

import { MCPServer, text } from "mcp-use/server";
import { z } from "zod";

const server = new MCPServer({
  name: "my-server",
  version: "1.0.0",
});

server.tool({
  name: "get_weather",
  description: "Get weather for a city",
  schema: z.object({ city: z.string() }),
}, async ({ city }) => {
  return text(`Temperature: 72°F, Condition: sunny, City: ${city}`);
});

await server.listen(3000);
// Inspector at http://localhost:3000/inspector

Step 3: Test with the MCP Inspector

After server.listen(), open http://localhost:3000/inspector in your browser to trace tool calls, replay requests, and debug MCP traffic without needing a full AI client setup.

Step 4: Deploy to Manufact Cloud

Connect your GitHub repo to Manufact MCP Cloud for production deployments with built-in metrics, latency tracking, and branch-based previews.

Python support follows the same pattern:

pip install mcp_use

from mcp_use.server import MCPServer, text
from pydantic import BaseModel

server = MCPServer(name="my-server", version="1.0.0")

@server.tool(name="get_weather", description="Get weather for a city")
async def get_weather(city: str) -> str:
    return text(f"Temperature: 72°F, Condition: sunny, City: {city}")

server.listen(3000)

Deeper Analysis

What makes this different from raw MCP SDK?

The official MCP SDK from Anthropic is low-level. mcp-use adds:

• Inspector-first debugging — every server ships with a built-in UI for tracing and replay
• MCP Apps — widgets that render as rich UI components inside ChatGPT and Claude, not just text responses
• Cross-client testing — run the same test suite across ChatGPT, Claude, and other MCP clients via Manufact Cloud
• Python-first SDK — the official Python MCP SDK is separate; mcp-use provides an opinionated Python DX equivalent

Pricing

The mcp-use SDK is open source and free (MIT license). Manufact Cloud has a free tier for personal projects and paid tiers for teams with branch deployments and analytics. The SDK itself does not require a Manufact account.

Practical Evaluation Checklist

• GitHub: 10,134 stars, active main branch (last push 2026-06-21)
• TypeScript SDK on npm: mcp-use
• Python SDK on PyPI: mcp_use
• Inspector available online at inspector.mcp-use.com or locally
• Supports both MCP servers (for agents) and MCP apps (for ChatGPT/Claude widgets)
• MIT license, no commercial restrictions on SDK usage

Security Notes

• mcp-use SDK has no special privileges on your system — tools run in your server process
• For Manufact Cloud deployments, credentials are managed via GitHub OAuth and per-deployment environment variables
• The MCP Inspector debugs traffic locally; production traffic on Manufact Cloud can be traced with the Cloud Inspector tool

FAQ

Q: Is this the same as the official Anthropic MCP SDK? A: No. The official MCP SDK from Anthropic is at github.com/modelcontextprotocol/sdk. mcp-use is a higher-level framework by Manufact that wraps the MCP protocol with an opinionated DX layer, Zod/Pydantic schema support, an Inspector UI, and MCP App widget rendering.

Q: Can I self-host without using Manufact Cloud? A: Yes. The SDK itself is fully self-hostable — server.listen(port) runs your MCP server on any HTTP endpoint. Manufact Cloud is an optional deployment and hosting layer.

Q: Does it support Python? A: Yes. The Python SDK is on PyPI as mcp_use and mirrors the TypeScript API with Pydantic instead of Zod.

Q: What clients support MCP Apps? A: ChatGPT and Claude are the primary targets for MCP Apps (the widget rendering feature). MCP Servers built with mcp-use work with any client that supports the MCP protocol, including Claude Code and custom agents.

Conclusion

mcp-use fills the gap between “I know what MCP is” and “I want to build, test, and ship an MCP server without pain.” The Inspector alone saves hours of debugging, and the TypeScript-first DX with Python parity makes it accessible to full-stack teams. If you are building any MCP-powered tool, the SDK is worth trying before reaching for the raw protocol.

• Docs: mcp-use.com/docs
• GitHub: github.com/mcp-use/mcp-use
• Inspector demo: inspector.mcp-use.com