Install any skill in seconds. Free to start, no credit card required.
Get Started Free →**MANDATORY for ALL MCP server work** - mcp-use framework best practices and patterns. **READ THIS FIRST** before any MCP server work, including: - Creating new MCP servers - Modifying existing MCP servers (adding/updating tools, resources, prompts, widgets) - Debugging MCP server issues or errors - Reviewing MCP server code for quality, security, or performance - Answering questions about MCP development or mcp-use patterns - Making ANY changes to server.tool(), server.resource(), server.promp
| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -18% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -41% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 9% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -63% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -63% | 0% |
mcp-use is the full stack TypeScript framework for building MCP servers and MCP apps to plug-ins and Claude connectors
Use it to build fully typed MCP servers with tools, resources, prompts, 1 line oauth authentication adapters, middleware, production transports, a built-in Inspector, and headless tooling.
mcp-use has native first class support for MCP Apps with support for HMR both at the server and MCP App level to live preview the changes on clients like ChatGPT
Follow the guides at <https://docs.mcp-use.com/v2>.
Consult the TSDoc bundled in the installed package and the TypeScript source comments on disk. Inspect the project's installed mcp-use version, generated types, and existing ChatGPT-specific integrations before choosing APIs or patterns.
Put reusable agent workflows in skills/<name>/SKILL.md; the directory is served automatically, so normally omit the skills server option. Use skills: false to disable it or skills: { directory: "server-skills" } to override the project-relative directory. Keep supporting references, scripts, templates, and assets in the skill instead of inflating tool descriptions.
Other measured skills in the registry, with their headline benchmark lift.