Basic Memory exposes 3 tools via FastMCP HTTP transport. All tools have descriptions and documented parameter schemas. Tool names follow verb_noun convention (search, fetch, cat). However, the definitions have moderate quality gaps: parameter descriptions are present but some are generic or incomplete (e.g., 'Optional FastMCP context' for the context parameter across all tools); output schemas are not documented; the 'cat' tool combines multiple concerns (line selection, section slicing, token limiting) in one tool; and error handling guidance is absent. The server demonstrates competent schema usage but lacks the polish and documentation depth expected of A-grade tools. Baseline: average tool description is 194 chars; parameter descriptions average 72 chars. All three tools meet minimum thresholds but show inconsistent quality.

• Document the output schema for each tool. Example for 'search': 'Returns {"content": string, "sources": [{"title": string, "url": string}], "metadata": {"count": int}}'. This enables LLMs to plan chaining.
• Refactor 'cat' into focused tools: 'get_note_content' (basic read), 'get_note_section' (section extraction), and 'get_note_excerpt' (with line/token limits). Each tool does one thing, making agent reasoning clearer.
• Replace the generic 'context' parameter with explicit, optional parameters: 'user_id' (if auth is required) or 'api_key' (if inline auth is needed). Do NOT pass opaque context objects, LLMs cannot reason about them.
• Add error handling to descriptions: 'search: Returns empty results if no matches found. fetch: Returns 404-style error if document not found, try search() first. cat: Fails if identifier is ambiguous, use memory:// URL or full project/path to disambiguate.'
• Document identifier precedence for 'cat': 'Resolves identifiers in this order: (1) memory:// URLs, (2) project/path strings, (3) exact title matches. If ambiguous, returns error with suggestions.'
• Add parameter constraints: For 'start_line' and 'end_line', document the valid range (1 ≤ start_line ≤ end_line ≤ num_lines). For 'max_tokens', specify the valid range (e.g., 1 - 10000) and default behavior if omitted.