AEO + Agent-API layer for Next.js. Makes sites readable by AI and callable by agents. Exposes actions as MCP tools and site pages as MCP resources.
Six tools with solid naming (verb_noun pattern) and complete JSON schemas. Descriptions are present and actionable (avg 95 chars), exceeding the 10-char minimum. All tools are READ_ONLY with clear intent. However, parameter descriptions are sparse or missing in several tools (e.g., list_pages 'cursor' and 'limit' lack descriptions), and output schemas are not documented. Error handling guidance is absent. Composition is strong, tools chain well (search_pages → get_page, list_pages → get_page). No security issues detected (read-only, no secrets).
Read the full AI-ready Markdown for one site page. Pass a route returned by list_pages or search_pages.
Get the markdown content of a documentation page by its route.
List all available API methods (actions) with their names and descriptions.
List AI-readable site pages in deterministic route order. Use the returned cursor to continue browsing large sites.
Search documentation pages by keyword. Returns matching pages with title, summary, and route.
Search page titles, routes, summaries, topics, questions, entities, and content using deterministic local lexical ranking. Optionally filter results by locale.
Output schemas not documented. LLMs cannot plan downstream calls or extract fields without knowing response structure (e.g., does search_pages return 'route' or 'url'? Does list_pages return 'total_count' or 'has_more'?).
Parameter descriptions missing or incomplete. 'list_pages' cursor and limit parameters lack descriptions; 'list_api_methods' has empty properties object. LLMs cannot infer parameter semantics from names alone.
No error handling guidance. Tools lack recovery hints (e.g., 'Route not found. Call list_pages() to discover valid routes'). Agents cannot self-correct on failures.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 69 | 2026-07-28+ | v2 |
Pagination cursor semantics unclear. 'list_pages' accepts a cursor but description does not explain format, max length rationale (512), or how to detect end-of-results. Agents may misuse pagination.
Tool annotations missing. No readOnlyHint, destructiveHint, or idempotentHint declared. Modern MCP clients cannot infer safety properties without explicit annotations.