Setting up the Vercel MCP Server
Connect Vercel's official MCP server (beta) via add-mcp or manually to Claude Code, Cursor, and others — with OAuth sign-in and security notes.
- Skill Road
- Setting up the Vercel MCP Server
Published on 09.09.2026
The Vercel MCP Server is hosted: nothing is installed locally, and the MCP client connects directly to https://mcp.vercel.com.
Fastest setup with add-mcp
npx -y add-mcp https://mcp.vercel.com -g
The tool automatically detects installed AI clients and configures Vercel MCP for all of them. Without -y it asks for confirmation before each install; without -g it installs only for the current project instead of globally.
Setting it up manually in Claude Code
claude mcp add --transport http vercel https://mcp.vercel.com
Then run /mcp to start the OAuth sign-in.
Setting it up in Codex CLI
codex mcp add vercel --url https://mcp.vercel.com
Codex automatically opens the browser for authorization when adding the server.
Setting it up in Cursor, VS Code, and other clients
Cursor and VS Code provide one-click install links in the official documentation. For Claude.ai and Claude Desktop, Vercel can be added as a custom connector under Settings → Connectors with the URL https://mcp.vercel.com (available for Pro, Max, Team, and Enterprise users). For ChatGPT, developer mode must first be enabled under Settings → Connectors → Advanced settings before a new connector with OAuth authentication can be created.
Staying secure
Vercel recommends enabling human confirmation for tool execution rather than letting agents make deployment or project changes unsupervised. When using multiple MCP servers simultaneously, be particularly careful about prompt injection attacks: malicious content from another connected server could try to steer the agent into unwanted actions in Vercel.
Keeping the beta status in mind
Vercel MCP is still in beta and subject to Vercel's Public Beta Agreement. Feature scope and supported clients may therefore change; the official documentation always reflects the current state.
Common connection issues
If add-mcp fails with an error about an unrecognized client, manual setup via the corresponding claude mcp add or codex mcp add command usually works. If no browser window opens during authorization, check whether the client is blocking pop-ups or whether a displayed URL needs to be opened manually.
For teams with multiple Vercel projects
Anyone working across multiple Vercel teams or projects should carefully check which team or project is currently connected before making production deployment changes. Since the agent operates with the signed-in account's permissions, a mix-up could otherwise lead to changes on the wrong project.
A quick check of the active context before any major deployment action is far faster than rolling back an accidental change afterward.
Building this check into a standard checklist before agent-driven deployments is worth the small overhead.
Source: vercel.com/docs/agent-resources/vercel-mcp and vercel.com/docs/mcp/vercel-mcp/tools, checked on 2026-09-06.
Frequently asked questions
Do I need to install anything for the Vercel MCP Server?
No. The server is hosted by Vercel and connected via a URL. The add-mcp tool only installs a small configuration in your existing AI clients, not a server of its own.
What does the Vercel MCP Server cost?
Per the provider, the server can be used on all Vercel plans without separate additional costs and is currently in beta.
Can I use the server without a Vercel account?
Yes, public tools such as documentation search require no authentication. Project, deployment, and analytics tools require a Vercel account and an OAuth sign-in.
Which AI clients are supported?
Officially supported clients include Claude Code, Claude.ai, ChatGPT, Codex CLI, Cursor, VS Code with GitHub Copilot, Devin, Raycast, Goose, Windsurf, and Gemini Code Assist and Gemini CLI. For security reasons, only clients reviewed by Vercel.
How do I protect myself against prompt injection with multiple MCP servers?
Vercel recommends manually confirming tool execution and exercising particular caution when using multiple MCP servers simultaneously, since malicious content from another server could try to steer the agent into unwanted actions.