Exposing Vercel to your agent via MCP

The official Vercel MCP server runs at https://mcp.vercel.com. By connecting it to Claude Code, you can query your deployments, read your logs, and inspect your projects directly from the terminal, without leaving your editor or opening the dashboard.

This is the third MCP bridge endpoint the Blokby course covers: after seeing how Claude Code consumes MCP servers and how Supabase exposes one, you connect the final link here, on the Vercel side.

What the server exposes

Vercel MCP organizes its tools into two categories.

Public tools (no authentication required):

• search_documentation - search the official Vercel docs

Authenticated tools (require your Vercel account):

• list_teams, list_projects, get_project - explore your project structure
• list_deployments, get_deployment - status and metadata of your deployments
• get_deployment_build_logs - build logs for diagnosing a failure
• get_runtime_logs - runtime logs filterable by level (error, warning, info), environment (production, preview), HTTP code, and time range
• check_domain_availability_and_price, buy_domain - domain management
• deploy_to_vercel - trigger a deployment from the agent

Adding the MCP server

The following command registers the Vercel server in Claude Code for the current project.

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

By default, the scope is local (current project only). If you want this server available in all your projects, add --scope user to the command above.

You can verify the server appears in your list:

$claude mcp list

Authenticating the server

The Vercel MCP server uses OAuth. The first connection opens an authorization flow in your browser.

Launch Claude Code in your project:

$claude

Then, once in the chat interface, display the MCP server status:

$/mcp

Claude Code opens your browser so you can authorize access to your Vercel account. Once confirmed, the server moves to "connected" status and its tools appear in the list.

First test: listing your deployments

Once connected, you can ask Claude Code to interact with Vercel in plain language. Try this prompt in Claude Code:

List my 5 most recent Vercel deployments and tell me which ones are in production.

Claude calls list_teams then list_projects to identify your project, then list_deployments to fetch the history. You can also query logs directly:

Show me the runtime error logs for the blokby-website project from the last hour.

Diagnosing a failed deployment

The most common use case once MCP is connected: understanding why a build failed without opening the dashboard.

My last deployment failed. Fetch the build logs and explain what happened.

Claude calls list_deployments to identify the failed deployment, then get_deployment_build_logs to read its logs. It then explains the error and can suggest a fix in the same conversation thread.