Cloudflare MCP Server

Official

Workers, KV, R2 and D1 by conversation — Cloudflare's official remote servers for building and observability.

Remote serverssestreamable-httpTypeScriptApache-2.0 2.9k

What is the Cloudflare MCP server?

Cloudflare went all-in on remote MCP earlier than almost anyone — they built some of the first OAuth-native hosted servers and the tooling others use to ship their own. The result is a family of official servers, each a URL away: documentation, Workers bindings (KV, R2, D1, queues), observability/logs, GraphQL analytics, DNS analytics, a browser-rendering server and more.

For developers on the Workers platform, the bindings server is the daily driver. Your assistant can list and inspect Workers, read and write KV namespaces, query D1 databases, and manage R2 buckets — the operational loop of edge development without the dashboard. Pair it with the observability server and debugging becomes conversational: "why is the api-proxy worker erroring?" pulls actual logs and error patterns rather than inviting speculation.

The docs server deserves separate mention as the zero-risk entry point: it searches current Cloudflare documentation (no account access needed), grounding answers about a platform that evolves weekly. Given how much of Workers' API surface is newer than any model's training data, that grounding pays for itself immediately.

Everything authenticates via OAuth against your Cloudflare account — no API tokens in config files — and each server is scoped to its job, so you connect only the capabilities you want an assistant holding. For teams shipping on the edge, this is the reference example of how a platform should do MCP: hosted, scoped, official, and boring to set up.

Rollout advice

Cloudflare publishes more than a dozen separate servers, so the real question is which subset to connect, not whether to.

  • Start with docs and observability. Both are effectively read-only and they answer the two questions that come up most: how does this API behave, and why is this Worker unhappy right now.
  • The bindings server writes. It can put KV keys, delete R2 objects and execute D1 statements, so aim it at a staging account before a production one.
  • Each server holds its own OAuth grant, revocable individually from the Cloudflare dashboard. Connecting five servers is five separate approvals, not one blanket handover.
  • Clients without native remote transport bridge through npx mcp-remote, which runs the OAuth flow and caches the resulting token under your home directory.

What you can do with it

Edge debugging

Worker errors diagnosed from real logs in conversation.

Data operations

KV, D1 and R2 inspected and modified without dashboard round-trips.

Platform-current answers

Workers APIs move fast; the docs server keeps answers honest.

Available tools

ToolWhat it does
workers_list / workers_get_workerEnumerate and inspect Workers scripts
kv_* toolsRead and write KV namespaces
d1_* toolsQuery D1 SQL databases
r2_* toolsManage R2 buckets and objects
(observability) query_logsSearch Workers logs and error patterns
(docs) search_docsSearch current Cloudflare documentation

How to install the Cloudflare MCP server

claude mcp add --transport sse cloudflare-docs https://docs.mcp.cloudflare.com/sse

Docs need no account. Add bindings/observability servers the same way for account access (OAuth prompt).

Configuration

A Cloudflare account for account-scoped servers (OAuth). The docs server works with no account at all.

Example prompts to try

  • List my Workers and tell me which ones haven't been deployed in six months.
  • Query the D1 database for the sessions table schema and row count.
  • Why is the image-resizer worker throwing 1101 errors? Check the logs.

Frequently asked questions

They're split by job: docs (no auth), bindings (Workers/KV/R2/D1), observability (logs), analytics and others. Add the two or three matching your workflow rather than everything.