Render MCP Server

The Render MCP Server is the official Model Context Protocol integration for Render, letting AI clients create and manage cloud infrastructure on Render through natural language. It exposes around 20+ tools covering workspaces, services (web services, static sites, cron jobs), deploys, logs, metrics, Postgres, and Key-Value instances. Render maintains the project under the render-oss GitHub organization and publishes the server as a Go binary, a Docker image, and a hosted streamable HTTP endpoint.

The hosted version at https://mcp.render.com/mcp is the recommended entry point: clients that support streamable HTTP MCP (Cursor, Claude Code, Codex, Claude Desktop with a connector, Windsurf, Jules, VS Code, Zed, Gemini CLI, Warp, Crush) can connect directly without running anything locally, and Render automatically rolls out new capabilities. Clients that only support stdio can run the binary or Docker image locally instead.

Authentication uses a Render API key (created from Account Settings in the Render Dashboard), passed as an Authorization: Bearer header for the hosted server or as the RENDER_API_KEY environment variable for local execution. The server is intentionally scoped: it can create resources and read everything, but it currently cannot trigger manual deploys, change scaling, delete resources, set IP allowlists, or create image-backed services. Postgres queries are read-only by design.

Tools

Prerequisites

• A Render account
• A Render API key from Account Settings → API Keys. Render API keys are broadly scoped and grant access to all workspaces and services on the account.

Option 1: Hosted server (recommended)

The official hosted endpoint is https://mcp.render.com/mcp. Authenticate with a Bearer token.

Claude Code (CLI):

claude mcp add --transport http render https://mcp.render.com/mcp \
  --header "Authorization: Bearer <YOUR_API_KEY>"

Cursor / Claude Desktop (HTTP MCP config):

{
  "mcpServers": {
    "render": {
      "url": "https://mcp.render.com/mcp",
      "headers": {
        "Authorization": "Bearer <YOUR_API_KEY>"
      }
    }
  }
}

OAuth is not yet supported; clients authenticate via the Authorization header.

Option 2: Local stdio (Docker)

{
  "mcpServers": {
    "render": {
      "command": "docker",
      "args": ["run", "-i", "--rm", "-e", "RENDER_API_KEY", "mcp/render"],
      "env": {
        "RENDER_API_KEY": "<YOUR_API_KEY>"
      }
    }
  }
}

Workspace selection

Render API keys can access multiple workspaces. After connecting, use list_workspaces and select_workspace, or instruct the assistant which workspace to operate in.

• Self-service deploys from chat: Spin up a new Node or Python web service on Render from a GitHub repo without leaving Cursor or Claude Code.
• Postgres exploration with an LLM: Run ad hoc read-only SQL against a production Render Postgres database to investigate data anomalies or generate reports.
• Incident triage: Pull recent error logs and CPU/memory/latency metrics for a misbehaving service, filtered by status code or path, to narrow down a regression.
• Env var management: Update API keys or feature flags on an existing service through a prompt, without opening the Render dashboard.
• Infra inventory: List every service, cron job, Postgres, and Key-Value instance across a workspace and summarize plan tiers and regions.

• "Create a new Node web service on Render called api-staging from the main branch of my acme/api repo, on the starter plan in Oregon."
• "Show me the last 50 error-level logs from my web service in the last hour, filtered to status code 500."
• "Run SELECT count(*) FROM users WHERE created_at > now() - interval '7 days' against my prod-db Postgres instance."
• "What were CPU and memory usage on worker for the last 24 hours? Plot it at 5 minute resolution."
• "Add STRIPE_API_KEY=sk_live_... to the checkout service env vars."

• Official and actively maintained by Render under the render-oss GitHub org, with regular releases.
• Hosted endpoint available at https://mcp.render.com/mcp, so most clients need zero local setup and get capability updates automatically.
• Broad surface area: workspaces, services, deploys, logs, metrics, Postgres, and Key-Value covered in one server.
• Direct read-only SQL against Render Postgres is genuinely useful for LLM-driven data exploration.

• Limited write surface: cannot trigger manual deploys, change scaling, delete resources, set IP allowlists, or create image-backed services. Updates on existing services are limited to environment variables.
• API key auth only: OAuth is not supported yet, and Render API keys are broadly scoped to the whole account.
• Metrics window capped at 30 days, and Postgres queries are read-only (no DDL or writes).

• Fly.io MCP server for similar PaaS-style deploys and management on Fly.
• Railway MCP server for managing services and databases on Railway.
• Cloudflare MCP servers for Workers, D1, R2, and other Cloudflare infrastructure.