Single tool 'build_research_brief' has a well-structured input schema with proper type definitions, constraints (minLength, maxItems, minimum/maximum), and clear descriptions for most parameters. Tool name uses verb_noun pattern correctly. Description is substantive (145 chars) and explains the deterministic, traceable nature. Output schema is documented with required fields. However, parameter descriptions lack depth on expected formats and use cases. The 'sourcePolicy' object parameter has no description of its structure or purpose. Error handling is present but generic. Tool annotations (readOnlyHint, idempotentHint) are correctly declared.

• Add detailed description to 'sourcePolicy' parameter: 'Optional metadata object describing the source's licensing, access restrictions, or attribution requirements. Expected fields: license (string), accessLevel (string), attribution (string).'
• Expand 'queries' description: 'Array of search queries (1-20) to match against document content. Queries are matched using full-text search; results ranked by relevance. Each query can return up to maxEvidencePerQuery evidence items.'
• Document the output schema structure in the tool description or as inline comments: 'Returns {brief: {title, question, summary}, evidence: [{query, source, excerpt, confidence}], run: {executedAt, documentCount, queriesProcessed}}'.
• Enhance error messages with recovery guidance: Instead of 'documents cannot contain more than 200 records', return 'documents array exceeds limit of 200. Split into batches of ≤200 and call build_research_brief multiple times, then merge results.'
• Add parameter 'maxEvidencePerQuery' description: 'Maximum evidence items to extract per query (1-10). Higher values increase comprehensiveness but may include lower-confidence matches. Default: 3.'.