REST API + MCP Server running on Cloudflare Workers. Provides product management tools via Model Context Protocol.
CF-MCP provides 5 tools with explicit schemas and descriptions visible in src/mcp.ts. All tools follow verb_noun naming convention (list_products, get_product, create_product, update_product, delete_product). Descriptions are present (avg ~60 chars) and convey intent clearly. However, several critical gaps prevent a higher score: (1) No output schemas documented, responses are inferred from REST API but not specified in tool definitions. (2) No error handling guidance or recovery instructions in descriptions. (3) Parameters lack format constraints, ranges, and validation guidance (e.g., 'price' is numeric but no min/max specified). (4) No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear READ_ONLY, WRITE, DESTRUCTIVE risk levels provided. (5) Descriptions lack WHEN to use each tool, making selection between similar tools unclear. (6) No mention of pagination or result limits, yet list_products could return hundreds of items. Average tool score: 62 across 5 tools.
Create a new product
Delete a product by ID
Get details of a specific product by ID
List all products with optional filtering by category or stock status
Update an existing product
No output schemas documented for any tool. Responses are inferred from REST API but not formally specified in tool definitions. LLMs cannot plan downstream chaining without knowing what fields to expect.
No error handling guidance in tool descriptions. When a product is not found, deleted, or creation fails, tools offer no recovery steps. LLMs have no actionable next step.
Tool annotations missing. Risk levels (READ_ONLY, WRITE, DESTRUCTIVE) are provided in metadata but not converted to tool annotations (readOnlyHint, destructiveHint, idempotentHint). This prevents clients from displaying safety warnings or enforcing permissions.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 54 | <=2025-11-25 | v2 |
| 2026-03-09 | D | 54 | 2024-11-05+ | v1 |
Descriptions are too brief and lack WHEN/WHY guidance. Average description ~55 chars; production baseline is 194 chars. 'Create a new product' does not explain when to call this vs updating an existing product, or what defaults apply.
No parameter constraints. 'price' is numeric but lacks min/max (prevent negative?). 'category' is free-form string but should be enum of {Electronics, Home, Sports} from examples. LLMs will hallucinate invalid values.
list_products lacks pagination. No limit, offset, or cursor parameters documented. If the product database grows to hundreds of items, LLM receives all in one response, exhausting context window and degrading reasoning.
Destructive operations (delete_product) lack confirmation/dry-run pattern. No mention of reversibility or warning to agents. Agents may accidentally delete products without confirmation.
Category parameter in list_products and create_product has example values in description ('e.g., Electronics, Home, Sports') rather than formal enum constraint. LLMs may assume these are the only valid values or may pass other values expecting them to work.