Cloudflare’s official MCP server. It exposes the Cloudflare API surface — Workers, KV namespaces, R2 buckets, Pages projects, DNS records, cache purges, analytics — as structured tools. The agent can deploy a Worker, flip a feature flag in KV, or purge a stale cache without you opening the dashboard.
What it produces: roughly 40 tools spanning the main Cloudflare products. workers_deploy, kv_get/put, r2_list_objects, pages_create_project, dns_list_records, cache_purge. All authenticated via API token.
Best for: anyone running 500k.io-style stacks (Pages + Workers + R2 + KV). Also useful for ops automation: “purge all /journal/* URLs from cache after deploy,” “rotate this DNS record,” “list all Workers using more than 10ms median CPU.”
Skip if: you’re on Vercel or Netlify — it’s Cloudflare-specific. Skip if you only use Cloudflare for DNS, the friction of an extra MCP isn’t worth it for one record-edit per quarter.
Setup gotchas: create a scoped API token (Account-level, with the minimum permissions for what you actually need — never the global key). Token in env: CLOUDFLARE_API_TOKEN. Cloudflare’s permission system is granular; “Edit Workers” doesn’t grant “Edit Pages.” Pre-test the token in wrangler whoami before wiring it into the MCP.
Real-world workflow: post-deploy, I tell Claude “purge cache for all changed paths in the last commit.” It pulls the diff, maps to URLs, calls cache_purge once. Used to be a 5-minute manual process.
Compatible alternatives: the wrangler CLI is still the right tool for local dev — this MCP shines for ad-hoc ops and deploy automation.
If 500k.io is on Cloudflare, this server lives in your config.