Vercel MCP 🔌 MCP server Freemium

What it does

Vercel MCP is Vercel's official Model Context Protocol server, currently in public beta. It is a hosted, remote server at https://mcp.vercel.com that uses OAuth, so there is nothing to install or run locally: your coding agent signs in to your Vercel account and can then work with your projects directly. The typical use is debugging from inside the editor. When a deployment fails, the agent can read the build and runtime logs, check the project settings and suggest a fix without you switching to the dashboard.

Tools it gives your agent

• Search and read Vercel documentation (available without signing in)
• List and inspect teams, projects and deployments
• Read deployment build logs and runtime logs
• Query Web Analytics: visitors, page views and custom events
• Deploy code; Vercel's changelog also announces support for purchases

The full list, split into public and authenticated tools, is in Vercel's tools reference.

Works with

Vercel only allows AI clients it has reviewed and approved. The documented list includes Claude Code, Claude.ai and Claude Desktop, ChatGPT, Codex CLI, Cursor, VS Code with Copilot, Devin, Raycast, Goose, Windsurf, Gemini Code Assist and Gemini CLI. Other MCP clients may be refused until Vercel adds them.

How to connect

In Claude Code, add the server and then sign in with /mcp:

claude mcp add --transport http vercel https://mcp.vercel.com

Vercel also documents npx add-mcp https://mcp.vercel.com, which detects your installed agents and configures each one.

Cost and limits

Using the MCP server does not cost extra, but it works on your Vercel account, so the usual Vercel plan limits and billing apply to anything the agent deploys or buys. The server is still in beta, and its tools and behaviour may change.

Safety notes

Connecting gives the agent the same access as your Vercel user account. Vercel asks for explicit consent for each client connection, to protect against confused-deputy attacks. It also warns about prompt injection: a malicious instruction in a web page or file could ask the agent to copy private deployment logs elsewhere. Keep human confirmation on for any change, and be especially careful before letting an agent deploy code or buy something.