mcp-use – TypeScript & Python MCP framework from Manufact

TL;DR

TL;DR: mcp-use is Manufact’s open-source MCP framework (TypeScript + Python) for building servers and cross-client apps, with an online inspector for debugging and a managed cloud for one-command production deployment.

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 API license.spdx_id)
• Latest release: [email protected] (published 2026-07-08, verified via GitHub Releases API)
• HN launch thread: news.ycombinator.com/item?id=48762862
• Stars: 10,283 (GitHub API, as of 2026-07-13)

What Is mcp-use?

The official description from the README:

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

Unlike a bare MCP SDK, mcp-use bundles three things:

1. SDK — TypeScript (npm install mcp-use) and Python (pip install mcp_use) packages
2. Inspector — a browser-based debugging UI at http://localhost:3000/inspector when the server runs
3. Cloud deployment — one-command push to Manufact MCP Cloud, or self-host via the @mcp-use/cli

Setup Workflow

Step 1: Install the SDK

TypeScript:

npx create-mcp-use-app@latest
# OR
npm install mcp-use

Python:

pip install mcp_use

Step 2: Build your first MCP server

TypeScript (from the 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:

from mcp_use import MCPServer, text
from zod import z

server = 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 def get_weather(city: str) -> str:
    return text(f"Temperature: 22°C, Condition: sunny, City: {city}")

server.listen(3000)

Step 3: Debug with the Inspector

The inspector runs in your browser at http://localhost:3000/inspector when the server is active. It lets you invoke tools, inspect requests and responses, and validate your schema definitions without needing a full MCP client.

The inspector is also available as a standalone open-source tool at github.com/mcp-use/mcp-use/tree/main/libraries/typescript/packages/inspector.

Step 4: Deploy to production

Option A — Manufact MCP Cloud (managed):

Connect your GitHub repo at manufact.com. Manufact handles branch-deployments, observability, metrics, and logs. No Dockerfile or CI config needed.

Option B — Self-hosted via CLI:

npx @mcp-use/cli login
npx @mcp-use/cli deploy

Deeper Analysis

MCP Apps vs MCP Servers

mcp-use differentiates between two artifact types:

• MCP Servers — traditional tool/Resource/provider definitions for AI agents
• MCP Apps — interactive widgets that render across multiple MCP clients (Claude, ChatGPT, etc.) from a single codebase

For MCP Apps, you define a React widget alongside the server. The widget receives props generated from the tool call and renders a rich UI — write once, display everywhere.

Skills for AI coding agents

mcp-use ships a skill for Claude Code, Codex, and Cursor that lets an AI coding agent scaffold a full MCP App project with the correct structure and conventions.

Python SDK parity

The Python SDK (mcp_use on PyPI) mirrors the TypeScript API surface, including tool registration, server lifecycle, and the inspector integration. Zod is used for schema validation in both languages.

Practical Evaluation Checklist

• TypeScript SDK: npm install mcp-use → works
• Python SDK: pip install mcp_use → works
• Inspector: http://localhost:3000/inspector loads when server runs
• MCP Apps: widget + server pattern confirmed in README
• Deploy CLI: @mcp-use/cli available on npm
• Manufact Cloud: GitHub repo integration described on manufact.com
• License: MIT (open-source, permissive)

Security Notes

• The MCP inspector endpoint (/inspector) binds to localhost by default. Do not expose it to the public internet in production.
• When deploying to Manufact MCP Cloud, review their data handling terms for any proprietary code you host.
• MCP Apps that render third-party widgets inherit the security model of the MCP client rendering them (Claude, ChatGPT, etc.).

FAQ

Q: How does mcp-use differ from the official Anthropic MCP SDK? A: The official SDK provides the low-level transport and protocol layer. mcp-use wraps it with a high-level API (typed schemas via Zod, tool decorators, MCP App widget support), an in-browser inspector, and a cloud deployment pipeline.

Q: Can I self-host without using Manufact Cloud? A: Yes. Use the @mcp-use/cli to deploy to your own infrastructure. Manufact MCP Cloud is optional — the framework and CLI are fully open-source.

Q: Does mcp-use support streaming and batch operations? A: Check the official docs at mcp-use.com/docs for the latest capability matrix, as streaming support evolves between releases.

Q: What is the current version? A: The latest release is [email protected] (published 2026-07-08 on GitHub Releases).

Conclusion

mcp-use turns the MCP protocol into a practical fullstack workflow: scaffold and debug locally in the browser inspector, then push to a managed cloud or your own server with one command. The dual TypeScript/Python SDK means you can use whichever language your stack already runs in. With 10k+ GitHub stars and active development (last push 2026-07-13), it is the most mature open-source MCP framework outside the official SDK.

If you are building AI agents that need to expose tools to multiple clients, mcp-use handles the boilerplate so you can focus on your tool logic.

Source: github.com/mcp-use/mcp-use | mcp-use.com