Model Context Protocol server for Cloudflare AutoRAG that provides search and aiSearch tools for AutoRAG instances
The server exposes two search-related tools with reasonable descriptions and partially complete schemas. However, parameter descriptions are sparse, output schemas are not documented, and error handling guidance is minimal. The tools follow a verb_noun naming pattern (search, aiSearch) but lack clarity on when to use one vs. the other. Schema completeness is moderate, required fields are marked, but nested object properties lack type annotations and descriptions. Tool descriptions are present but generic (78-92 chars), missing dependency hints or recovery guidance.
Search through Cloudflare AutoRAG instance documents and get AI-powered responses
Search through Cloudflare AutoRAG instance documents
Missing output schema documentation. Input schemas are defined, but no documentation of what fields the response contains, their types, or structure. LLMs cannot plan downstream operations without knowing what to expect.
Incomplete parameter descriptions. The nested 'ranking_options' object has only a generic description; its 'score_threshold' property lacks context on valid ranges, units, or when to use it. 'max_num_results' has no guidance on limits or default behavior.
No distinction between search and aiSearch. Both tools accept nearly identical parameters and return similar results. The descriptions do not explain when to use one vs. the other ('AI-powered responses' is vague). This violates the pattern that similar tools must have obviously distinct names.
Inferred effective spec: 2025-06-18+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 53 | 2025-06-18+ | v2 |
| 2026-03-09 | F | 31 | 2025-03-26+ | v1 |
No error handling guidance. If a query returns no results, or ranking_options are invalid, or the Cloudflare instance is unreachable, the tool provides no recovery path. Error responses should guide the LLM (e.g., 'No results. Try broadening your query or lowering score_threshold.').
Missing numeric constraints. 'max_num_results' is defined as a number with no minimum, maximum, or default. An LLM might pass 0, 10000, or -1, causing undefined behavior. Should declare: max 1 - 1000, default 20.
Undocumented pagination. aiSearch accepts a 'cursor' parameter but tool description does not explain how pagination works, what cursor values are valid, or when to call it. Is cursor opaque? What does it return? This leaves the LLM guessing.
No mention of result limits or truncation. If a search matches thousands of documents, does the tool return all of them? Returns 1000? 100? A cap should be documented and enforced. High token counts risk context exhaustion.