Summary

The Figma and Magic connectors are placeholders — read that before planning around them.

The README is explicit: those files contain placeholders and need real API calls and MCP client logic written before the integration tools do anything. What works today is the Vercel AI SDK side. The intended pattern is orchestration in the editor — call the Figma server, then the Magic server, then this one — rather than server-to-server calls, and the composite tool that would do it directly is present in the code but not the supported path.

What it is

An MCP server wrapping core Vercel AI SDK functions so an editor assistant can call them alongside other servers, rather than reimplementing structured generation in the editor.

What you get

Tools wrapping Vercel AI SDK Core functions, including `generate_object` for structured output and `generate_ui_component`
`set_tool_category`, a meta-tool that switches which category of tools is active — a way to keep the number of exposed tools inside what Cursor handles well
A deployment path to Smithery: a `Dockerfile` and `smithery.yaml` ship with the project, and API keys are supplied at deploy time
Both transports covered — stdio locally, or SSE when `TRANSPORT_TYPE` says so

Requirements

Node.js v20 or later, npm and git; the project is vercel-ai-sdk-mcp-project, version 1.0.0. Clone it, `npm install`, `npm run build`, then `npm run start`. `OPENAI_API_KEY` is required for the AI SDK tools; `ANTHROPIC_API_KEY` is optional. `FIGMA_API_KEY` and `TWENTY_FIRST_API_KEY` are only needed once the corresponding connectors are implemented. Locally the config points at `dist/index.js`; deployed, Cursor runs it through the Smithery CLI.

Vercel AI SDK MCP Server Project