Vercel operates an official remote MCP at mcp.vercel.com with streamable HTTP and documented tools for Vercel projects — searching documentation, listing teams and projects, inspecting deployments, streaming build and runtime logs, and helping with rollbacks. Complete OAuth in your client, then let the model answer operational questions the same way you would in the dashboard.
https://mcp.vercel.com
Claude Sonnet 4.5
MCP Playground runs 30+ models on the same workflow: switch anytime, or use Compare mode to run several in parallel and balance quality vs. cost.
Vercel account via OAuth in supported clients, or a Vercel personal token where your MCP client allows bearer auth (vercel.com/account/tokens).
How models use it and what it is built for.
Vercel MCP is aimed at “agentic DevOps” on top of Vercel: the model can look up a topic in Vercel’s docs, list recent deployments, fetch failed build logs, read runtime errors, check domain and DNS state, and (with the right tool and confirmation) create deployments or use safe CLI helpers. Some tools are public, others require an authenticated Vercel user. The server is the same one Vercel documents for first-party clients like Codex, Claude, and Cursor, so you get predictable behaviour in MCP Agent Studio once OAuth or headers are set up in your client.
Typical tools an AI model can call. Exact names vary by version.
Copy any of these into MCP Agent Studio after connecting.
Why did the last production deploy fail? Pull the build log summary.
List all environment variable keys (not values) for project “acme” on my team.
What preview URLs exist for the current branch, and is any deployment marked ERROR?
Promote the latest preview to production and confirm the production alias.
This is not a single-model product: you get the same MCP connection with 30+ models (Claude, GPT, Gemini, DeepSeek, open-weight, and more), you can switch mid-conversation, and you can open Compare mode to run the same prompt against multiple models at once. The card above is a suggested starting point for this server — not the only choice.
Default pick for Vercel
Claude Sonnet 4.5
Claude Sonnet 4.5 is strong on log triage, multi-file deploy reasoning, and safe rollback suggestions. For doc-only Q&A, GPT-5.2 or Haiku 4.5 is often enough.
Open MCP Agent Studio with the connection pre-filled. Add your token, pick any of 30+ models, and start chatting — no install required.
Try Vercel in Agent StudioCommon questions about connecting, scoping and using it safely.
It is the hosted Model Context Protocol service Vercel runs at mcp.vercel.com. It exposes Vercel product APIs as well-documented tools so an LLM can search docs, list projects, inspect deployments, read logs, and work with domains according to the published tools reference.
Use the Vercel template so the base URL and transport are pre-filled. In clients that need OAuth, complete the Vercel sign-in flow when the connector prompts. If your client only accepts a bearer key, use a Vercel personal access token with the least scope required.
Some tools are write-capable, such as creating deployments or buying domains, depending on the tool list for your account. Start with a team or project that is safe to experiment on, and confirm any destructive or spend-related action before the client sends it to Vercel’s API.
Vercel documents which first-party and partner clients they test with, but the protocol is standard. Any MCP client that can speak streamable HTTP and complete OAuth (or use a valid token) can call the same endpoint — including MCP Playground in the browser.
Use Claude Sonnet 4.5 or a similar “reasoning” tier so the model can correlate build logs, configuration, and documentation. You can A/B in Agent Studio next to a cheaper model to see whether the extra cost buys fewer wrong hypotheses.