Vercel MCP Pro
helbertparanhos/vercel-mcp-pro
Controle total da Vercel via MCP com 70 ferramentas e modo readonly
Ver no GitHubServidor MCP completo para a Vercel com 70 ferramentas cobrindo toda a REST API (deployments, projetos, variáveis de ambiente, domínios, DNS, certificados, logs). Oferece modo somente leitura para segurança e escape hatch com vercel_raw para cobertura 100% da API.
Instalação
npx convez-ai helbertparanhos/vercel-mcp-pro1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Automatizar deployments e gerenciar projetos Vercel via Claude
- Consultar logs, variáveis de ambiente e configurações de DNS programaticamente
- Integrar pipeline CI/CD com IA para gestão de infraestrutura Vercel
- Implementar escape hatch com vercel_raw para operações customizadas
Veredito

Implementação robusta e bem mantida que supera a solução oficial Vercel em cobertura de API (70 ferramentas vs. funcionalidades limitadas). Recomendado para times que precisam orquestração avançada de infraestrutura Vercel com IA, especialmente com modo readonly para segurança em ambientes sensíveis.
README
A tradução para o português deste README ainda está na fila.
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_rawescape 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
- Create a token at https://vercel.com/account/settings/tokens.
- Copy
.env.example→.envand 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=trueblocks every write/delete tool — only reads andvercel_rawGETs run. Ideal for audits and exploration.- The
vercel_rawpath is sanitized (must start with a version segment like/v9/…; no scheme/host/traversal/control chars) and constrained toapi.vercel.com. - Auth, retries (429 with
Retry-After, 5xx, network) and rate-limiting are handled for you. The token only ever travels in theAuthorizationheader — never in a query string or log.
Security model — read this before trusting the guards
VERCEL_READONLYblocks mutations, not reads of secrets. Read tools likelist_env_vars(withdecrypt:true),get_env_varandget_edge_config_itemsreturn 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_IDis a convenience scope, not a security boundary. It sets the default team, but a caller can pass a differentteamId/slugper call (including viavercel_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