MCP server for Cloudflare Workers, Pages, and KV. List Workers, inspect deployments, manage KV, query metrics — all from Claude Desktop, Cursor, or any MCP client.
Server has 9 tools with basic schemas and descriptions, but significant gaps prevent higher scores. All tools have descriptions (10-100 chars, mostly adequate), and input schemas are present with Zod validation. However, parameter descriptions are minimal or absent, most params lack explanatory text beyond the name. Output schemas are undocumented; responses are raw JSON dumps without field documentation. No error handling guidance, no pagination support despite list tools, and no tool annotations (readOnlyHint/destructiveHint). Naming is verb-first and clear (list_, get_, kv_), which is good. The server follows basic MCP patterns but lacks production-grade polish.
Metadata + bindings for one Worker script.
Get the value for one key in a KV namespace.
Set the value for one key in a KV namespace. Expiration TTL in seconds optional.
All Workers KV namespaces.
Recent deployments with build status and preview URLs.
All CF Pages projects with their custom domains.
Parameter descriptions missing or minimal. Most params (e.g., 'name', 'zone_id', 'key') have only a short label, not explanatory text. LLMs cannot infer whether 'name' is a display name, script name, or internal identifier without explicit guidance.
Output schemas undocumented. Tools return raw Cloudflare API responses (JSON.stringify) without field documentation. LLMs cannot plan downstream calls or extract relevant data without knowing response structure.
No pagination support on list tools. list_workers, list_pages_projects, list_kv_namespaces accept no limit/offset parameters. Large result sets will blow context windows and degrade LLM reasoning.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 57 | 2026-07-28+ | v2 |
Routes binding URL patterns to Workers in one zone.
All Worker scripts on the account.
Aggregated request count, error count, and CPU time over the last 24 hours.
No error recovery guidance. Error messages are raw HTTP status + text slice (e.g., 'CF GET /accounts/...: 401 ...'). LLMs receive no actionable next steps, should suggest 'Check CF_API_TOKEN' or 'Verify CF_ACCOUNT_ID'.
No tool annotations. kv_put is destructive (WRITE risk) but lacks destructiveHint annotation. list_* tools are read-only but lack readOnlyHint. Agents cannot distinguish safe from risky operations without explicit hints.