Vercel MCP Pro

Vercel MCP Pro ▲

The most complete Vercel MCP server — 70 tools covering the entire Vercel REST API (deployments, projects, env vars, domains, DNS, aliases, certs, logs, checks, webhooks, edge config, teams) plus a vercel_raw escape hatch and a readonly safety mode.

Works in any MCP client — Claude Code, Claude Desktop, Cursor, ChatGPT — using a Vercel access token (no OAuth flow required). Ships with an optional /vercel skill that drives the Vercel CLI for terminal deploys.

Why this and not the official one?

	Official mcp.vercel.com	vercel-mcp-pro
Auth	OAuth (approved clients only)	Token (works everywhere, self-hosted)
Coverage	Docs search, manage projects/deployments, analyze logs	Full REST API: env, domains, DNS, aliases, certs, webhooks, edge config, checks, log drains…
Escape hatch	—	vercel_raw → any endpoint, 100% coverage
Safety	—	VERCEL_READONLY=true blocks all writes

Install

npx -y vercel-mcp-pro

Or clone and build locally:

git clone https://github.com/helbertparanhos/vercel-mcp-pro
cd vercel-mcp-pro
npm install && npm run build

Configuration

1. Create a token at https://vercel.com/account/settings/tokens.
2. Copy .env.example → .env and fill it in:

Variable	Required	Description
VERCEL_TOKEN	✅	Personal or team access token.
VERCEL_TEAM_ID	—	Default team scope (injected as ?teamId). Leave empty for personal account.
VERCEL_TEAM_SLUG	—	Scope by team slug (alternative to VERCEL_TEAM_ID).
VERCEL_READONLY	—	true blocks all write/delete tools (and only allows vercel_raw GETs).
VERCEL_DISABLE_RAW	—	true removes the vercel_raw escape hatch.
VERCEL_TIMEOUT_MS	—	Request timeout (default 60000).
VERCEL_MAX_RETRIES	—	Retries on 429/5xx/network (default 3).

Add to your client

Claude Code

claude mcp add vercel-mcp-pro -- npx -y vercel-mcp-pro

Or in this project's .claude/settings.json → mcpServers:

"vercel": {
  "command": "node",
  "args": ["projects/vercel-mcp-pro/dist/index.js"],
  "env": {
    "VERCEL_TOKEN": "your_token",
    "VERCEL_TEAM_ID": "team_xxx"
  }
}

Claude Desktop

%APPDATA%\Claude\claude_desktop_config.json (Windows) / ~/Library/Application Support/Claude/claude_desktop_config.json (Mac):

{
  "mcpServers": {
    "vercel": {
      "command": "npx",
      "args": ["-y", "vercel-mcp-pro"],
      "env": { "VERCEL_TOKEN": "your_token", "VERCEL_TEAM_ID": "team_xxx" }
    }
  }
}

Cursor

Paste the same config into .cursor/mcp.json.

Tools (70)

Deployments (11)

list_deployments · get_deployment · create_deployment · cancel_deployment · delete_deployment · get_deployment_events · list_deployment_files · get_deployment_file_contents · promote_deployment · rollback_deployment · get_promote_aliases

Projects (7)

list_projects · get_project · create_project · update_project · delete_project · pause_project · unpause_project

Environment Variables (6)

list_env_vars · get_env_var · create_env_var · bulk_create_env_vars · update_env_var · delete_env_var

Domains (9)

list_domains · get_domain · get_domain_config · add_domain · verify_domain · remove_domain · list_project_domains · add_project_domain · remove_project_domain

DNS (4)

list_dns_records · create_dns_record · update_dns_record · delete_dns_record

Aliases (4)

list_aliases · get_alias · assign_alias · delete_alias

Certificates (4)

get_cert · issue_cert · upload_cert · remove_cert

Logs (2)

get_build_logs · get_runtime_logs

Checks (5)

list_checks · get_check · create_check · update_check · rerequest_check

Webhooks (3)

list_webhooks · create_webhook · delete_webhook

Edge Config (6)

list_edge_configs · get_edge_config · get_edge_config_items · create_edge_config · update_edge_config_items · delete_edge_config

Teams & User (4)

get_user · list_teams · get_team · list_team_members

Integrations & Log Drains (4)

list_integrations · list_log_drains · create_log_drain · delete_log_drain

Escape hatch (1)

vercel_raw — call any Vercel REST endpoint (method + full versioned path + params + body). Guarantees 100% coverage even for niche/new endpoints (marketplace, sandboxes, feature-flags, access-groups, rolling-release…). Auto-injects your team scope; honored by readonly mode for GETs.

Common recipes

Diagnose a failed deploy

list_deployments(projectId:"my-app", state:"ERROR")  →  get_build_logs(idOrUrl:"dpl_...")

Ship a preview to production (no rebuild)

promote_deployment(projectId:"my-app", deploymentId:"dpl_...")

Roll back fast

rollback_deployment(projectId:"my-app", deploymentId:"dpl_previous")

Add an env var to all environments

create_env_var(projectId:"my-app", key:"API_KEY", value:"…", target:["production","preview","development"])

Anything not covered

vercel_raw(method:"GET", path:"/v1/security/firewall/config", params:{ projectId:"my-app" })

The /vercel CLI skill (bundled)

For terminal-driven deploys, this repo also ships a Claude Code skill in skill/ that drives the official Vercel CLI (vercel deploy, vercel logs, vercel env, vercel rollback). The MCP gives full API coverage in any client; the skill gives the smoothest git push → deploy loop in the terminal. They complement each other — see skill/SKILL.md.

Safety

• VERCEL_READONLY=true blocks every write/delete tool — only reads and vercel_raw GETs run. Ideal for audits and exploration.
• The vercel_raw path is sanitized (must start with a version segment like /v9/…; no scheme/host/traversal/control chars) and constrained to api.vercel.com.
• Auth, retries (429 with Retry-After, 5xx, network) and rate-limiting are handled for you. The token only ever travels in the Authorization header — never in a query string or log.

Security model — read this before trusting the guards

• VERCEL_READONLY blocks mutations, not reads of secrets. Read tools like list_env_vars (with decrypt:true), get_env_var and get_edge_config_items return decrypted values and still run in readonly mode — that's their purpose. If your environment must never surface secrets to the model, don't expose those tools / run against a least-privilege token.
• VERCEL_TEAM_ID is a convenience scope, not a security boundary. It sets the default team, but a caller can pass a different teamId/slug per call (including via vercel_raw). The real boundary is the token's own scope — use a token limited to the team/projects you intend to automate.
• Treat the token as production credentials. It grants the same access as your Vercel account/team. Prefer a scoped token, keep it in .env (gitignored) or your client's secret store, and rotate it if it's ever shared or pasted into a chat.

License

MIT © Helbert Paranhos / Strat Academy