mcp-server-cloudflare
This repository contains a collection of MCP servers built by Cloudflare. MCP stands for Model Context Protocol, which is a standard that allows AI tools like Claude or Cursor to talk to external services and take actions on your behalf. Think of it as a bridge: instead of manually logging into a dashboard to check logs, adjust settings, or look up data, you can type a question in plain English to an AI assistant, and it reaches out to the appropriate service and handles it for you.
Cloudflare runs a wide range of internet infrastructure services, including security, content delivery, serverless computing, DNS, and analytics. This repository provides separate MCP servers for different parts of that platform. Each server is focused on a specific area, such as viewing application logs and analytics, managing serverless worker deployments, running internet traffic analysis, fetching and screenshotting web pages, querying audit logs, or checking DNS performance. There are 14 servers listed in the README, each with its own hosted URL.
Connecting one of these servers to an MCP-compatible client is mostly a matter of pointing the client at the right URL. If the client supports remote MCP servers natively, you paste in the server URL directly. If not, there is a setup step involving a configuration file and a small bridge package. The README includes a short code snippet showing what that configuration looks like.
Cloudflare also maintains a separate, broader MCP server in a different repository called the Code Mode server, which covers many Cloudflare APIs through general code execution rather than purpose-built tools. The README explains the tradeoff: the servers in this repository are better when you are focused on one product area and want guided, structured interactions, while the Code Mode server suits workflows that need to span many products at once.
Some features require a paid Cloudflare Workers subscription. The README also notes a known issue where very large AI responses can get cut off due to context length limits, and it suggests keeping queries short and splitting complex requests into smaller ones.