Vercel MCP Server

Official

Deployments, logs and projects from Vercel's official hosted server — ship and debug without the dashboard.

Remote serverstreamable-httpTypeScriptProprietary

What is the Vercel MCP server?

Vercel's official MCP server at mcp.vercel.com gives assistants a hosted, OAuth-authenticated line into your Vercel account: projects, deployments, build and runtime logs, plus documentation search — the operational surface of shipping on Vercel, minus the dashboard tabs.

The deploy-debug loop is the core value. "Why did the last production deployment fail?" walks the assistant into the actual build logs, where the real error (the type mismatch, the missing env var, the ESM/CJS incident) lives — read, interpreted and, if your coding assistant has repo access, often fixed in the same conversation. That loop — deploy fails, assistant reads logs, assistant patches code, redeploy — is one of the tightest agentic workflows currently available anywhere.

Beyond firefighting, it answers the fleet-management questions that accumulate around an account: which projects exist, what domains point where, which deployments are on which branches, what changed between Friday's working deploy and Monday's broken one. Docs search grounds platform questions in current Vercel documentation, which matters for a platform iterating as quickly as this one.

Being an official hosted server, setup is one URL plus an OAuth approval, and access rides your Vercel login with its existing team scoping. Client support is broad (Claude, Cursor, VS Code, Devin and friends). If your deploys land on Vercel, this plus a repo-connected coding assistant is the shortest path from "build failed" to "fixed and shipped" that exists right now.

Notes from the field

  • Long builds produce long logs, and long logs get truncated. If the first read comes back inconclusive, ask for the failing step by name instead of the whole transcript.
  • Team scoping catches people out. Call list_teams first when you belong to several, or a personal-account default will hand back a confusingly empty project list.
  • Runtime log retention is short on lower plans, so "why did it 500 last Tuesday" often has no answer left to find. Reach for the runtime tools while the incident is still warm.
  • The pairing is what makes it good. Alone it explains failures; with repository access in the same thread it edits next.config.js, commits, and watches the redeploy go green.

What you can do with it

Deploy-failure autopsy

Build logs read and interpreted the moment a deployment goes red.

Fleet questions

Projects, domains and deployment states across an account, answered directly.

Agentic ship loop

With repo access alongside: fail → read logs → patch → redeploy, in one thread.

Available tools

ToolWhat it does
list_projects / get_projectEnumerate and inspect projects
list_deployments / get_deploymentDeployment history and status
get_deployment_build_logsRead build logs — the deploy-failure goldmine
get_runtime_logsRuntime/function logs for live debugging
search_documentationSearch current Vercel docs
list_teamsTeam context for scoped operations

How to install the Vercel MCP server

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

OAuth approval in the browser; access follows your Vercel account and teams.

Configuration

A Vercel account; OAuth handles the rest. No local install for HTTP-capable clients.

Example prompts to try

  • Why did the latest production deploy of the marketing site fail?
  • List my projects and flag any without a production deployment this month.
  • Read the runtime logs for the api project — what's causing the 500s since yesterday?

Frequently asked questions

OAuth against your Vercel account — approve once in the browser, revoke anytime from account settings. Team access mirrors what your login can already see.