The Cloudflare API MCP server at mcp.cloudflare.com/mcp is a remote service that exposes the Cloudflare platform through a small “Code Mode” tool surface: the model plans calls against a machine-readable API spec, then executes the right Cloudflare API requests. Connect with OAuth in supported clients, or with a least-privilege API token in automation-friendly flows such as Agent Studio.
https://mcp.cloudflare.com/mcp
Claude Sonnet 4.5
MCP Playground runs 30+ models on the same workflow: switch anytime, or use Compare mode to run several in parallel and balance quality vs. cost.
OAuth in supported UIs, or a Cloudflare API token (My Profile → API Tokens) with least scope; paste the bearer into Agent Studio for token-based clients.
How models use it and what it is built for.
Cloudflare’s hosted MCP wraps thousands of account-scoped API operations — Workers and Pages, DNS, KV, D1, R2, load balancing, Zero Trust, and more — without hand-writing one integration per product. The default Code Mode server minimises the tool list; you can also disable it with a `?codemode=false` query to register one tool per API route if you prefer traditional fixed schemas. The same product rate limits and permission model apply as the Cloudflare API you already use. Point any MCP client at the URL, authenticate, and the model can help deploy, inspect, or debug infrastructure in natural language.
Typical tools an AI model can call. Exact names vary by version.
Copy any of these into MCP Agent Studio after connecting.
List all Workers in my account and the routes each one serves.
Show the DNS A and CNAME records for my production zone and flag anything unusual.
What R2 buckets do I have and which look empty or unused?
Is there a recent spike in 5xx for my production Worker, and in which colos?
This is not a single-model product: you get the same MCP connection with 30+ models (Claude, GPT, Gemini, DeepSeek, open-weight, and more), you can switch mid-conversation, and you can open Compare mode to run the same prompt against multiple models at once. The card above is a suggested starting point for this server — not the only choice.
Default pick for Cloudflare
Claude Sonnet 4.5
Claude Sonnet 4.5 handles multi-step infra reasoning and high-risk API plans. For narrow read-only checks, Haiku 4.5 is cheaper; compare in Agent Studio.
Open MCP Agent Studio with the connection pre-filled. Add your token, pick any of 30+ models, and start chatting — no install required.
Try Cloudflare in Agent StudioCommon questions about connecting, scoping and using it safely.
It is Cloudflare’s remote MCP endpoint that connects AI tools to the Cloudflare API. In the default “Code Mode” design, the model searches the published API specification and issues the right `cloudflare.request` style calls, so you can manage Workers, DNS, R2, and other services through chat without custom glue code for every product.
Preferred for humans is OAuth when your client can open a browser. For headless and browser labs, create an API token in the Cloudflare dashboard with only the account permissions the agent should have, and pass it as a bearer token with the MCP request.
Treat it like any high-privilege API client: use a token scoped to a narrow set of products, start on staging or a throwaway account, and review proposed destructive actions. You can also split “read” and “write” by using a read-only token for exploration.
Yes — any MCP client that can reach the hosted URL and complete OAuth or pass a token can work. In MCP Playground, open Agent Studio, pick the Cloudflare template, and paste the URL and token as prompted.
For multi-step change plans (DNS plus Workers, comparing regions, parsing errors), use Claude Sonnet 4.5. For one-off “list my buckets” style reads, try Claude Haiku 4.5. Compare models in Agent Studio on a real change before you standardise.
Vercel
List deployments, read logs, manage env keys and roll back from natural language.
GitHub
Drive GitHub repos, PRs and issues with an AI agent.
Supabase
Manage Supabase projects, run SQL and inspect schemas with AI.
Sentry
Triage issues, stack traces and releases from chat without opening ten browser tabs.