Cloudflare MCP Server

Overview

Cloudflare MCP Server exposes the Cloudflare platform APIs as a suite of Model Context Protocol tools, letting agents deploy and manage Workers, query and mutate KV and R2 storage, create and update DNS records, run D1 SQL databases, and invoke Workers AI models — all without leaving an agent session. Hosted as a remote MCP service by Cloudflare, it requires no local server process; agents connect with a scoped API token over HTTP/SSE transport. It works best with Claude Desktop, Claude Code, and any MCP-compatible host that supports remote servers, including Cursor and Windsurf. The server is the reference example of a major cloud vendor shipping a first-party, production-grade MCP surface, making it an essential integration for teams building or operating anything on the Cloudflare edge.

Best for

Best for
Teams building or operating applications on the Cloudflare edge who want agents to drive deployments, storage, DNS, and AI workloads end to end

Use cases

• Deploy a Cloudflare Worker, bind a KV namespace, and set a DNS route in a single agent session
• Inspect and rotate R2 bucket contents or KV keys through a conversational agent
• Manage DNS records across multiple zones without leaving the agent loop
• Run D1 SQL queries and Workers AI inference from an agent workflow
• Automate Cloudflare configuration drift detection and remediation

Notes

What it does

Cloudflare MCP Server translates the Cloudflare platform REST APIs into MCP-native tools that any compatible agent can call directly. The primary tool groups cover Workers (deploy, update, tail logs), KV (read, write, list, delete keys), R2 (list buckets, get/put objects), D1 (run SQL queries against serverless SQLite databases), DNS (create, update, delete records across zones), and Workers AI (invoke hosted AI models). Because it runs as a hosted remote MCP endpoint operated by Cloudflare, there is no local binary to install or keep updated.

How to deploy

Point your MCP host at Cloudflare’s remote MCP endpoint and supply a Cloudflare API token with the scopes your agent needs. In Claude Desktop, add the server under mcpServers in claude_desktop_config.json with the remote URL and your token. For Claude Code or Cursor, use the equivalent remote server configuration. Cloudflare recommends creating a dedicated token per project with the minimum required permissions rather than using a global token.

Best practices

Scope API tokens tightly — create one token per agent project and restrict it to the zones and products that project actually touches. Pair with the GitHub MCP server for a full code-to-deploy loop: the agent can pull a PR, build the Worker bundle, and deploy it to Cloudflare without leaving the session. Monitor Workers tail logs through the MCP surface to close the feedback loop after deployment.