Cloudflare Documentation MCP Server

The Cloudflare Documentation MCP Server is Cloudflare's official Model Context Protocol server for its own developer documentation. Unlike the Cloudflare API MCP Server, also listed in the Skill Road catalog, which authenticates via OAuth or an API token against your Cloudflare account and can write to DNS, Workers, R2, or Zero Trust rules, this server is a pure read-only tool for publicly available documentation content. According to the official repository README, it requires no authentication at all — any MCP-capable client can connect directly without a Cloudflare account. The official repository lives at https://github.com/cloudflare/mcp-server-cloudflare (Apache-2.0 license, exactly 4,210 stars per the GitHub API on 2026-09-19, actively maintained, last pushed 2026-09-01), with the corresponding product page at https://developers.cloudflare.com/docs-for-agents/.

What is the Cloudflare Documentation MCP Server?

Technically, this is a hosted remote MCP server operated by Cloudflare itself at https://docs.mcp.cloudflare.com/mcp. According to the provider, it runs on top of a Cloudflare AI Search (formerly AutoRAG) instance indexed with the complete Cloudflare developer documentation. Rather than rigid full-text search, the server returns semantically relevant, context-aware excerpts for each query — whether about Workers pricing details, R2 bucket configuration, or Workers AI Search binding usage. For AI-assisted development environments, this reduces hallucinations according to the provider, since the agent doesn't rely on outdated model knowledge but instead retrieves current documentation content in real time. The server is one of now 16 product-specific MCP servers Cloudflare maintains in the same repository alongside the broader Code Mode server (the Cloudflare API MCP Server mentioned above) — each covering a single product area such as observability, DNS analytics, Radar, or documentation itself.

Features and tools

According to the README, the server exposes exactly one tool: search_cloudflare_documentation. With it, an agent can ask natural-language questions about the Cloudflare platform, such as "Do Cloudflare Workers costs depend on response sizes if I'm serving map tiles from an R2 bucket?" or "How many indexes are supported in Workers Analytics Engine, with an example using the Workers binding API?". Answers are drawn exclusively from the indexed Cloudflare documentation, not from general model knowledge. That makes the server especially useful inside coding assistants when developers need quick, reliable, up-to-date answers about Cloudflare products without switching to a browser.

Distinction from the Cloudflare API MCP Server

Skill Road's catalog already includes the Cloudflare API MCP Server (cloudflare-mcp-server), which exposes over 2,500 Cloudflare API endpoints via code execution and requires OAuth against your account. The Documentation server here is deliberately narrower: it only reads public documentation content, changes nothing in your Cloudflare account, and needs no account at all. For pure research tasks during development, it's the lower-risk, faster-to-set-up choice; once you actually need to read or change account configuration, the API server remains the right tool.

Setup

Since the server is publicly reachable at https://docs.mcp.cloudflare.com/mcp and requires no authentication, adding the URL is usually all it takes. For Claude Code: claude mcp add --transport http cloudflare-docs https://docs.mcp.cloudflare.com/mcp. Clients without native remote-MCP support can connect through the bridge npx -y mcp-remote@latest https://docs.mcp.cloudflare.com/mcp. The Cloudflare AI Playground (playground.ai.cloudflare.com) also accepts the server URL directly in its interface.

Security notes

Because the server only provides read access to publicly published documentation, requires no authentication, and processes no account data whatsoever, its risk profile is very low — there simply are no write operations and no credentials that could be compromised. It's worth noting that retrieved documentation excerpts are embedded into the request sent to whichever language model is in use; anyone working with a proprietary model backend should factor that into their model choice.

FAQ

Is the Cloudflare Documentation MCP Server free? Yes. According to the provider, the server is public and usable without authentication, and the source code in the repository is Apache-2.0 licensed. There are no separate costs for using the documentation search.

Do I need a Cloudflare account? No. Unlike the Cloudflare API MCP Server, the documentation search needs no account and no API token.

How does it differ from the Cloudflare API MCP Server? The API server offers write-capable, account-bound access to over 2,500 endpoints and requires OAuth. This Documentation server offers read-only access to public documentation content, with no sign-in at all.

How current are the GitHub stars? The GitHub API reported exactly 4,210 stars for the repository https://github.com/cloudflare/mcp-server-cloudflare on 2026-09-19. That's a point-in-time snapshot, not a quality judgement.

A compatible MCP client with remote MCP server support (Streamable HTTP); no Cloudflare account or authentication required.

For Claude Code: claude mcp add --transport http cloudflare-docs https://docs.mcp.cloudflare.com/mcp. Without native remote support: use npx -y mcp-remote@latest https://docs.mcp.cloudflare.com/mcp as a bridge. No sign-in required.

npx -y mcp-remote@latest https://docs.mcp.cloudflare.com/mcp

No authentication required — the server is publicly reachable and needs no Cloudflare account.

Read-only access to public Cloudflare developer documentation; no access to account, zone, or customer data.

Requests go through Cloudflare-hosted Streamable HTTP at docs.mcp.cloudflare.com/mcp; search internally uses Cloudflare AI Search (AutoRAG) over the indexed developer documentation. Responses are returned to the MCP client and its language model.

Very low risk profile: read-only access to public content, no authentication, no account or customer data involved. Retrieved documentation excerpts flow into model requests.