Vercel's official MCP server lets an AI client work with your Vercel account — list projects and deployments, read build and runtime logs, and help debug failed deploys. It's a hosted remote server. Here's the setup.
Setup — official remote server
Add https://mcp.vercel.com as a remote/OAuth connector:
{
"mcpServers": {
"vercel": {
"url": "https://mcp.vercel.com"
}
}
}
Authorise via Vercel OAuth on first use. For stdio-only clients, bridge with npx -y mcp-remote https://mcp.vercel.com. Restart the client; see MCP config file location.
What it gives your AI
Inspect projects and deployments, read logs to debug a failing build, and query deployment status — useful when paired with a coding agent fixing the underlying issue.
Verify
Ask: "Why did my last Vercel deployment fail? Pull the build logs and summarise the error." A real log-grounded answer confirms it works.
Common problems
- OAuth scope/team — authorise the correct Vercel team; re-run if you see the wrong projects.
- No write actions — the server focuses on read/inspect; deployment changes still go through git/CLI.
mcp-remoteerrors — Node 18+; see Cursor MCP not working.
Going further
Vercel pairs with GitHub (fix the code) and Sentry (find the error). Browse the devops category or curated loadouts. You can also deploy your own MCP server to Vercel.