Memex has 4 tools with basic schemas and descriptions, but significant gaps limit production readiness. All tools have descriptions (10-80 chars, mostly adequate) and input schemas with type definitions. However, output schemas are completely undocumented, LLMs cannot plan downstream calls or extract structured data. Parameter descriptions are minimal (e.g., 'The text to remember' for cognify.data). No error handling guidance, no pagination support for search results, and no tool annotations (readOnlyHint/destructiveHint). The cognify tool modifies state but lacks confirmation/dry-run patterns. search tool accepts top_k but no documented limit enforcement or pagination cursor. Overall structure is present but shallow, typical of early-stage community servers.

• Document output schemas for all tools. Example: search() should return {results: [{id, text, score, dataset}], total_count, next_cursor}. This enables agents to chain tools and extract required fields.
• Add tool annotations: mark cognify as destructiveHint=true (modifies state), cognify_status and search as readOnlyHint=true. This guides agent planning and safety checks.
• Expand parameter descriptions with constraints. E.g., cognify.data: 'Text to add (1-10000 chars, UTF-8, no binary)'; search.top_k: 'Number of results (1-100, default 20)'.
• Implement pagination for search: add limit (default 20, max 100), offset/cursor, and return total_count. Document in description: 'Returns up to 20 results; use offset for pagination.'
• Add error recovery guidance. E.g., cognify error response: 'Dataset not found. Call list_data() to see available datasets.' This enables agents to self-correct without manual intervention.

Inferred effective spec: <=2025-11-25.

• Relies on Stateful initialize / Mcp-Session-Id (removed; protocol is stateless) - make each request self-contained