mcp-use is a full-stack MCP framework designed to build MCP Servers and MCP Apps that work across AI agents and clients like ChatGPT and Claude. The project provides a TypeScript/JavaScript server API to define tools, and a React-based widget system for building interactive MCP Apps. Use it to expose a set of tools (endpoints) that can be queried by clients, and optionally attach UI widgets for rich, cross-client experiences. The included Inspector helps you test tools and interactions in real time, making it straightforward to verify schemas, responses, and widget behavior before deploying to production.

With the TypeScript path, you typically create a server instance, register tools with descriptions and schemas, and then start listening on a port. The built-in inspector lets you inspect tool calls and widget data during development. For production deployments, you can rely on the Deploy flow (via the MCP Cloud) to provide observability, metrics, logs, and branch-based deployments, ensuring your server and apps stay healthy in production environments.

Prerequisites:

• Node.js (LTS) and npm installed
• Basic TypeScript/JavaScript tooling (optional but recommended)

1. Install Node.js from https://nodejs.org and verify: node -v npm -v

2. Install the MCP Use package (example: create a new MCP Server project using the CLI): npx create-mcp-use-app@latest

   Or initialize an existing project by installing the dependency locally:

   npm install mcp-use

3. If you plan to run a built server directly: npm run build

   Then run the server, e.g.:

   node dist/index.js

4. Optional: install and run the Inspector during development: npx @mcp-use/inspector --url http://localhost:3000/mcp

5. For production deployment, follow the Deploy workflow: npx @mcp-use/cli login npx @mcp-use/cli deploy

Tips and common considerations:

• The mcp-use server supports both a tooling API and optional widgets. Tools have a name, a description, and a schema describing required inputs.
• Widgets are auto-discovered from resources/ and can be used to present results in a UI across clients.
• When configuring environments, expose PORT for local development and ensure your deployment target can reach the Inspector if needed.
• If migrating from a different MCP setup, ensure your tool schemas align with the zod-based schemas used in examples to guarantee proper validation.
• For production, use the Manufact MCP Cloud or equivalent observability stack to monitor metrics, logs, and performance. Ensure environment variables, secrets, and authentication are configured as required by your deployment.