# Generate an instant MCP server for your API Generate a hosted MCP server from your OpenAPI spec. Every operation becomes a tool that Claude Code, Cursor, and any other MCP client calls directly. ## Generated from your spec, hosted by Fern - **One tool per operation.** Every supported operation in your spec becomes a tool with a typed input schema. - **Curate the tools agents see.** Filter, group, and merge operations into fewer, clearer tools, so agents pick the right one. - **Add custom tools and hooks.** Write tools and request hooks in TypeScript for workflows no single operation covers. - **Hosted at your own URL.** Fern builds and runs the server at `.fernmcp.dev`, so there is no infrastructure to manage. - **Same URL on every release.** Regenerate with the same slug and the server updates in place, so agent configs keep working. - **Safe deploys.** If a deploy fails, the server keeps serving the previous deployment and agents keep working. ## From spec to agent in three steps 1. **Add the target.** Add a `language: mcp` target with `delivery: hosted` and a `slug` to `fern/sdk-config.yml`. 2. **Generate.** `fern generate` builds the server in the cloud and prints its URL. Run it in CI with `FERN_TOKEN` to redeploy on every spec change. 3. **Connect an agent.** Paste the connect command into Claude Code, Cursor, or VS Code. ## Every endpoint, a tool Each operation in your spec becomes a tool, and its parameters and request body become the input schema agents see in tools/list. Rename any tool with `x-fern-mcp-name` in an overlay. ## Connect any agent with one snippet - **Ready for any MCP client.** The dashboard's Connect button builds setup for Claude Code, Cursor, VS Code, raw JSON, or a plain URL. - **Your auth, passed through.** Agents send your API's own credential as a header. A missing one returns an error that names the header to add. ## Monitor it from the dashboard Every hosted server gets an overview with its connection URL, tools, generation history, and analytics. - **Roll back in one click.** Pick an earlier successful generation and Fern redeploys it at the same URL. - **See agent traffic.** Analytics shows how often agents call your server, starting with the first request. - **Pause without deleting.** Take the endpoint offline, then resume on the same deployment when you're ready. ## Get started - Book a demo: - Get started: - Manage in the dashboard: