Vercel MCP Server
Connect AI agents to Vercel — manage projects, deployments, domains, and logs via OAuth.
Updated
What is Vercel MCP?
The Vercel MCP server is a Model Context Protocol server that gives AI tools secure access to your Vercel projects, deployments, domains, and logs. It is Vercel's official, first-party server.
It runs as a remote, hosted MCP server at https://mcp.vercel.com — there's nothing to install or self-host. Connections authenticate with OAuth: you authorize the client (e.g. Claude Code, Cursor, ChatGPT) against your Vercel account through a consent screen, and the agent then operates with the same access as your Vercel user. Vercel only allows clients it has reviewed and approved to connect.
Through the server, an assistant can search Vercel documentation, list teams and projects, inspect deployments and their build/runtime logs, check and buy domains, manage Toolbar comment threads, and even trigger deployments. It implements the latest MCP Authorization and Streamable HTTP specifications.
Tools the Vercel MCP server exposes
| Tool | What it does |
|---|---|
| search_documentation | Search Vercel documentation for a given topic. |
| list_teams | List all Vercel teams the authenticated user belongs to. |
| list_projects | List Vercel projects for a team. |
| get_project | Get details for a project — framework, domains, and latest deployment. |
| list_deployments | List a project's deployments with creation time, state, and target. |
| get_deployment | Get details for a deployment — build status, regions, and metadata. |
| get_deployment_build_logs | Fetch a deployment's build logs to investigate failures. |
| get_runtime_logs | Fetch runtime logs for a project/deployment, filterable by level, status, time, and text. |
| check_domain_availability_and_price | Check whether domain names are available and their price. |
| buy_domain | Purchase a domain with registrant information. |
| get_access_to_vercel_url | Create a temporary shareable link to a protected deployment. |
| web_fetch_vercel_url | Fetch content from a (optionally protected) Vercel deployment URL. |
| list_toolbar_threads | List Vercel Toolbar comment threads for a team. |
| reply_to_toolbar_thread | Add a reply to an existing Toolbar comment thread. |
| deploy_to_vercel | Deploy the current project to Vercel. |
Connect the Vercel MCP server
Claude Code
- 1
Add the server
In your project run: claude mcp add --transport http vercel https://mcp.vercel.com
- 2
Start Claude Code
Run `claude` in the same directory.
- 3
Authenticate
Type /mcp and complete the Vercel OAuth consent flow in your browser to authorize access.
You can also run `npx add-mcp https://mcp.vercel.com` to auto-configure every detected client.
Cursor
- 1
Edit your MCP config
Add Vercel to your project or global .cursor/mcp.json: { "mcpServers": { "vercel": { "url": "https://mcp.vercel.com" } } }
- 2
Let Cursor connect
Cursor picks up the server and shows a 'Needs login' prompt.
- 3
Authorize
Click the prompt and complete the Vercel OAuth flow to grant access.
Sample use cases
A scheduled agent monitors recent deployments and pulls build/runtime logs when one fails.
Failing builds and production errors are surfaced and summarized automatically.
On a deploy webhook, an agent fetches the new deployment URL to smoke-check key pages and endpoints.
Broken deployments are caught before they reach users.
A developer asks an AI assistant to look up Vercel docs and project/domain details without leaving their editor.
Faster answers and project context inline, no context switching.
Security & permissions
Vercel MCP uses OAuth — there are no API keys to paste. Each client connection requires an explicit OAuth consent screen, and Vercel maintains an allowlist of approved clients plus confused-deputy protections so cached consent can't be reused by a malicious request.
Once authorized, the connected AI client has the same access as your Vercel user account across the teams and projects you belong to — including reading project metadata, build and runtime logs, and (where the tool allows) triggering deployments, buying domains, and editing Toolbar threads. Treat the connection accordingly: enable human confirmation for tool execution, be cautious running it alongside untrusted MCP servers (prompt-injection risk), and review what data each agent in a workflow can reach.
Using Vercel MCP with Gamut
In Gamut, you add Vercel as a connected tool and Gamut manages the OAuth connection, so an agent can call the Vercel MCP server inside automated, event- or schedule-triggered workflows. For example, a scheduled agent can poll list_deployments and pull get_deployment_build_logs or get_runtime_logs to detect failing builds or production errors and post a summary to chat; or a deploy-webhook-triggered run can fetch a deployment URL with web_fetch_vercel_url to smoke-check it. Because the server grants the same access as your Vercel account, keep agents scoped to the read/monitoring tools you intend and add human confirmation before any state-changing action (deploying, buying a domain, resolving Toolbar threads).
Frequently asked questions
What is the Vercel MCP server?
It's Vercel's official Model Context Protocol server, hosted at https://mcp.vercel.com, that lets AI tools like Claude and Cursor securely search Vercel docs and manage your projects, deployments, domains, and logs over an OAuth connection.
Is the Vercel MCP server free?
There's no separate charge for the MCP server — it's included with a Vercel account and works on the free Hobby tier. Your usage is governed by your existing Vercel plan, and paid actions like buying a domain bill to your account as normal.
Is the Vercel MCP server hosted or self-hosted?
It's a hosted, remote server run by Vercel at https://mcp.vercel.com. There's nothing to install or self-host — you just point an approved MCP client at the URL and authorize via OAuth.
How do I connect the Vercel MCP server to Claude Code or Cursor?
For Claude Code, run `claude mcp add --transport http vercel https://mcp.vercel.com`, then type /mcp to authenticate. For Cursor, add { "vercel": { "url": "https://mcp.vercel.com" } } under mcpServers in .cursor/mcp.json and click the 'Needs login' prompt. Both use Vercel's OAuth consent flow.