Vercel MCP is Vercel's official Model Context Protocol server — a hosted endpoint at https://mcp.vercel.com that uses OAuth so AI clients like Cursor, Claude, and VS Code Copilot can work with your Vercel account directly.
This is not the same as deploying your own MCP server to Vercel — this is Vercel's own server for managing your account. Everything below is taken directly from the official Vercel MCP tools reference.
💡 Before you start: If you want to explore any MCP server's tools interactively before configuring your IDE, you can use the free browser-based tester at MCP Playground — paste the URL, connect, and browse tools without touching a config file.
Quick setup
Install helper (any client):
npx add-mcp https://mcp.vercel.com
Cursor — .cursor/mcp.json:
{
"mcpServers": {
"vercel": {
"url": "https://mcp.vercel.com"
}
}
}
Open Cursor → connect the server → complete OAuth when prompted. Other supported clients: Claude Code, ChatGPT (connector), VS Code Copilot, Codex CLI — full list here.
1. Search Vercel docs from inside your editor
Tool: search_documentation
Instead of alt-tabbing to a browser, ask your assistant to search Vercel's docs directly. You pass a topic — edge functions, env vars, routing — and it returns grounded, current answers from Vercel's actual documentation with an optional text limit.
No more "I think Next.js does it this way" guesses from stale training data.
2. List your teams and projects
Tools: list_teams, list_projects
Most other tools need a team and project identifier. These two let the assistant resolve the correct target automatically before calling anything downstream — especially useful when you have personal and org projects mixed together.
3. Get detailed project info
Tool: get_project
Returns a structured snapshot for one project: detected framework (Next.js, Astro, etc.), attached domains, latest deployment summary, and other metadata the Vercel API exposes. Gives the model factual grounding for questions like "what framework is this app using?" or "what's live right now?"
4. Browse deployment history + inspect a specific deploy
Tools: list_deployments, get_deployment
list_deployments is your history view — states, targets, timing. get_deployment is the drill-down: build status, regions, and metadata for one deploy by ID or URL.
Typical flow: list to find the failing preview → get detail on that single deployment → pull logs.
5. Debug failed builds
Tool: get_deployment_build_logs
Retrieves build-phase logs (not runtime logs) for a deployment, with a configurable line limit. When your preview is red, the assistant can fetch the logs for that deployment ID and help you identify dependency errors, wrong build commands, or missing build-time env vars — without you navigating to the Vercel dashboard log viewer.
6. Filter and search runtime logs
Tool: get_runtime_logs
This is the strongest operational tool in the set. It surfaces runtime output from Vercel Functions — console.log, errors, request traces.
Documented filters:
- Environment (production or preview)
- Log level
- HTTP status
- Source type (serverless vs edge)
- Time range
- Full-text query
- Request ID
That combination is ideal for AI-assisted debugging when users are hitting 500s or you need to trace one specific bad request.
7. Check domain availability and buy domains
Tools: check_domain_availability_and_price, buy_domain
Check one or more domain names, then walk through buy_domain with registrant fields when you're ready. These are real commerce actions — treat them like any sensitive dashboard operation and verify before approving tool runs.
8. Work with protected deployments
Tools: get_access_to_vercel_url, web_fetch_vercel_url
Preview deployments are often protected. Two different tools for two different needs:
-
get_access_to_vercel_url— creates a temporary shareable link so someone can open a protected preview URL without disabling protection globally. -
web_fetch_vercel_url— fetches the actual response body from a deployment URL, including when Vercel Authentication is required. Good for verifying a page or API route returns the right output.
9. Deploy and use the Vercel CLI from your assistant
Tools: deploy_to_vercel, use_vercel_cli
-
deploy_to_vercel— deploys the current project through the MCP tool surface. -
use_vercel_cli— steers the model toward CLI-first patterns, including--helpfor specific commands when you prefer that workflow.
10. Use a project-scoped MCP URL
From the Vercel docs:
https://mcp.vercel.com/{team-slug}/{project-slug}
By pointing at a project-specific endpoint, the server already knows your team and project context. Fewer manual slug parameters, better errors when context is missing, and a cleaner workflow when one repo maps to one Vercel project.
See Advanced usage in the Vercel docs.
Real-world scenarios
| Scenario | Tools involved |
|---|---|
| Red preview build |
list_deployments → get_deployment_build_logs
|
| 500s in prod |
get_runtime_logs (filter by status + time) |
| "What's live?" |
list_projects → get_project
|
| Ship from the IDE | deploy_to_vercel |
| Review a protected preview |
get_access_to_vercel_url or web_fetch_vercel_url
|
| Buy a domain |
check_domain_availability_and_price → buy_domain
|
A note on security
Vercel explicitly recommends:
- Human confirmation before tool runs (don't let the model act autonomously on deployments)
- Reading their prompt injection and confused-deputy guidance when multiple MCP servers are connected
Worth 5 minutes of reading before you wire this up to an autonomous agent.
FAQ
Is Vercel MCP free?
The MCP server itself is free to use if you have a Vercel account. The actions it performs (deploys, domain purchases) use your account's normal plan limits and billing.
Does this work with Claude Desktop?
Vercel's supported client list includes Claude Code (CLI). Check Supported clients for the current list.
Is this the same as deploying my own MCP server to Vercel?
No. This article is about using Vercel's hosted MCP to manage your account. If you want to host your own MCP server on Vercel infrastructure, that's a separate guide — Vercel documents that separately.
References
Want to try MCP servers interactively before configuring your IDE? MCP Playground lets you connect to any remote MCP server from the browser, browse its tools, and run them with live JSON-RPC logs — free, no sign-up required.
Top comments (0)