Nancy Brain MCP Server has a moderate quality baseline. All 6 tools are properly named with action verbs (search_, retrieve_, explore_, set_, get_), and all have descriptions present. Input schemas are complete with type definitions and parameter descriptions. However, several critical gaps reduce the score: (1) Output schemas are not documented in the provided code, no return type specifications are visible for any tool, making it impossible for LLMs to plan downstream calls. (2) Error handling guidance is absent, no recovery hints, retryability classification, or actionable error messages are evident. (3) Tool descriptions, while present, are functional but not optimized for LLM prompt-engineering (they lack explicit WHAT/WHEN/WHY framing). (4) The write operation (set_retrieval_weights) lacks confirmation/dry-run capability despite modifying persistent state. (5) Parameter descriptions are present but generic, they do not include constraint details, ranges, or dependency hints. The tools follow good naming conventions and have basic schemas, placing this in the C+/B- range rather than higher.

• Document output schema for each tool. Include field names, types, and descriptions. Example for search_knowledge_base: {results: [{doc_id: string, title: string, score: number, excerpt: string}], total_count: integer, next_cursor?: string}
• Add error handling to each tool. Categorize errors as retryable (rate_limit, timeout) or user-fixable (not_found, invalid_input) and provide recovery guidance. Example: 'Document not found. Try searching with a broader query or exploring the tree first with explore_document_tree.'
• Add confirmation step to set_retrieval_weights. Either require explicit user approval before persisting, or implement a dry_run parameter that returns the computed weight change without applying it.
• Specify parameter constraints in descriptions. Add ranges/enums: 'threshold (0.0 - 1.0, default 0.0)', 'start (1-based line number, 1 - 999999)', 'max_depth (1 - 10, default 3)'.
• Rewrite tool descriptions with explicit WHEN/WHY. Example: 'Use search_knowledge_base to find documents by keyword. Returns ranked snippets. For precise extraction of known document sections, use retrieve_document_passage instead.'
• Document mutually exclusive parameters. In set_retrieval_weights, explicitly state: 'Provide EITHER doc_id (to weight a single document) OR namespace (to weight all documents in a category), not both.'
• Add pagination guidance to search_knowledge_base if result count can exceed limit. Include next_cursor field and explain how to fetch additional results.
• Specify what get_system_status returns (e.g., uptime, embeddings_loaded, cache_status, error_count) so agents know which fields to expect.