Cloudflare MCP Server: Deploy and Manage Workers With AI

Cloudflare's hosted MCP API server connects an AI client to Cloudflare operations. Start with the matching server listing, choose the intended account, and review permissions before asking for an infrastructure change. Use the cloud category to compare other providers without relying on undated star counts.

Docs reviewed October 17, 2026 from Cloudflare's MCP server guide and official repository. This guide is docs-derived; MCPFind did not authorize an account, inspect infrastructure, or deploy a Worker.

What Does the Cloudflare MCP Server Give You Access To?

Cloudflare documents API access across products such as Workers, DNS, R2, and Zero Trust, subject to the granted permissions. An agent's ability to discover an API operation does not authorize it to run that operation. Distinguish a resource lookup from deployment, deletion, or policy changes.

How Do You Connect the Cloudflare MCP Server to Claude or Cursor?

Use the hosted endpoint https://mcp.cloudflare.com/mcp with Streamable HTTP in a compatible client. Follow Cloudflare's current setup instructions and your client's remote-server configuration:

{
  "mcpServers": {
    "cloudflare-api": {
      "url": "https://mcp.cloudflare.com/mcp"
    }
  }
}

Complete OAuth and review the requested permissions for interactive use. For CI or automation, Cloudflare documents scoped user or account API tokens passed as Authorization bearer tokens. Store the token through protected secret configuration, not in a shared file. The previous local npx example and legacy /sse examples are not the default hosted setup.

How Do You Use Cloudflare MCP for D1, R2, and KV?

Cloudflare currently recommends its API server for most work; product-specific servers are optional. Do not infer a current Bindings endpoint or identical permissions from an older setup example. Check the documented server and its permissions for the product and account involved.

For a D1 or storage task, begin with metadata in a development account rather than asking the agent to seed, overwrite, or delete data. Compare the returned resource identity with the Dashboard before broadening the task.

What Cloudflare Workers Tasks Can AI Agents Handle Through MCP?

A bounded first task is an inventory check:

In the named account, find this known Worker and summarize its deployment information. Return the resource identity and the evidence used. Do not deploy, change routes, or modify any resources.

The expected result identifies the intended account and Worker. Compare it with the Dashboard. If access fails, review account selection and token scope before granting broader permissions.

For a later deployment, have the agent present the target, code diff, expected effect, and rollback plan. Approve that mutation explicitly after review. Apply the same boundary to DNS, R2, D1, and policy changes; permission to inspect a resource is not approval to change it.

How Do You Deploy Your Own MCP Server on Cloudflare Workers?

Consuming Cloudflare's API MCP server is a different task from hosting your own tools on Workers. Use the current Cloudflare Agents documentation for the appropriate server, transport, authentication, and state pattern. The remote MCP hosting guide provides a comparison starting point; do not treat an old helper name or edge-location count as a deployment specification.

How Does Cloudflare MCP Compare to Other Cloud Infrastructure Servers?

Choose the server for the infrastructure you actually manage and the permissions it can constrain. A Cloudflare connection does not grant access to another provider. For a repeated deployment with an existing review and audit path, keep that workflow and evaluate MCP first for inspection or planning. Review product limits and costs before authorizing new resources.