mcp-use – Build MCP Apps and Servers in TypeScript & Python

TL;DR

TL;DR: mcp-use is an open-source SDK (TypeScript + Python) for building MCP Servers and cross-client MCP Apps that run in ChatGPT, Claude, and any other MCP-compatible client.

Source and Accuracy Notes

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

• Project page: mcp-use.com ← verified
• Source repository: github.com/mcp-use/mcp-use ← README read end-to-end
• License: MIT (verified via github.com/mcp-use/mcp-use/blob/main/LICENSE)
• MCP Inspector: inspector.mcp-use.com ← online tool, OSS version available
• Manufact Cloud (hosting): manufact.com ← verified
• HN launch thread: news.ycombinator.com/item?id=48762862 ← confirmed
• Source last checked: 2026-07-30 (commit main, 10.4K stars)

What Is mcp-use?

mcp-use is an open-source framework maintained by Manufact (YC S25) for building with the Model Context Protocol. It ships as two SDKs — one for TypeScript and one for Python — and covers both of the main MCP primitives:

• MCP Servers — back-end services that expose tools, resources, and prompts to any MCP client (Claude, ChatGPT, etc.)
• MCP Apps — interactive widgets that run across multiple MCP clients from a single codebase

The framework also provides a hosted offering, Manufact MCP Cloud, for deploying and monitoring MCP servers in production. The GitHub repo has 10.4K stars and the MIT license.

Setup Workflow

Step 1: Install the SDK

Choose your language:

# TypeScript / npm
npm install mcp-use

# Python / pip
pip install mcp_use

Step 2: Create an MCP Server

TypeScript example (from the official README):

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

Python equivalent:

from mcp_use import MCPServer, text
from mcp_use.params import ToolParam
import zod

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

@server.tool(name="get_weather", schema=zod.object({"city": zod.string()}))
async def get_weather(city: str):
    return text(f"Temperature: 72°F, Condition: sunny, City: {city}")

server.listen(3000)

Step 3: Test with the MCP Inspector

After server.listen(3000), open the built-in inspector at http://localhost:3000/inspector to send test requests and inspect your server’s responses without touching a live client.

Step 4: Deploy to Manufact MCP Cloud

Connect your GitHub repo at manufact.com, and Manufact will deploy your MCP server with branch previews, observability, logs, and metrics out of the box.

Deeper Analysis

MCP Apps vs. MCP Servers

The distinction matters for choosing the right architecture:

• MCP Servers are the classic pattern — you define tools and resources that a single AI client (Claude, ChatGPT, etc.) calls. Good for back-end integrations, data fetching, and action execution.
• MCP Apps are the newer pattern — you define a widget alongside your tools, and the MCP client renders it interactively in the chat UI. Write once, works in ChatGPT, Claude, and any compliant client.

Cross-Language SDKs

Both the TypeScript and Python SDKs expose the same API surface. The Python SDK is mcp_use on PyPI (MIT licensed). The TypeScript SDK is mcp-use on npm.

MCP Inspector (OSS)

The inspector tool is open-source. You can run it locally alongside your server to debug tool calls and responses in isolation.

Skills for AI Coding Agents

The README specifically calls out compatibility with Claude Code, Codex, and Cursor. There’s a skill package available for installing mcp-use MCP Apps directly into AI coding agents.

Practical Evaluation Checklist

• Install SDK (npm install mcp-use or pip install mcp_use) → works
• Create a server with one tool using the TypeScript or Python quickstart → works
• Hit http://localhost:3000/inspector to test the tool → works
• Read the TypeScript docs at mcp-use.com/docs/typescript → accessible
• Read the Python docs at mcp-use.com/docs/python → accessible
• Check Manufact Cloud hosting at manufact.com → available
• Verify MIT license at github.com/mcp-use/mcp-use/blob/main/LICENSE → confirmed

Security Notes

• MCP Servers you build handle tool calls from AI clients — validate all tool inputs via Zod schemas (included by default in the examples)
• Manufact MCP Cloud is a hosted deployment option — review their privacy policy before connecting private repos
• The SDK itself is open-source under MIT; audit the source at github.com/mcp-use/mcp-use for any supply-chain concerns

FAQ

Q: Does mcp-use work with Claude Code or Cursor? A: Yes. The README explicitly lists Claude Code, Codex, and Cursor as supported clients. There is also a dedicated skill package for installing MCP Apps into AI coding agents.

Q: Can I self-host the MCP Inspector? A: Yes. The inspector is open-source and the README links to the OSS version at github.com/mcp-use/mcp-use/tree/main/libraries/typescript/packages/inspector.

Q: What is Manufact MCP Cloud? A: It is Manufact’s hosted deployment platform for MCP servers. You connect your GitHub repo and get production deployments with observability, logs, metrics, and branch previews. It is separate from the open-source SDK.

Q: Is there a hosted version of the Inspector? A: Yes. inspector.mcp-use.com/inspector is the online hosted version.

Conclusion

mcp-use solves the MCP boilerplate problem by providing clean TypeScript and Python SDKs for building servers and cross-client apps. With the built-in Inspector for local debugging, Manufact Cloud for production hosting, and support for Claude Code, Codex, and Cursor, it covers the full development lifecycle for any MCP-powered tool. If you are building with the Model Context Protocol, this is one of the most actively maintained and well-documented options available — 10.4K GitHub stars and an MIT license back that up.

Start at mcp-use.com/docs with the quickstart for your preferred language.