18 tools with complete schemas and descriptions, but quality is uneven. All tools are READ_ONLY with good naming (verb_noun pattern). Descriptions range from 50 - 150 chars, meeting the 10 - 1024 baseline. Schemas present for all tools with proper types and defaults. However, descriptions lack actionable context (when to use vs similar tools, prerequisites, error recovery). Parameter descriptions are minimal (e.g., 'Minimum confidence tier' without explaining what confidence measures). No output schema documentation visible. Error handling guidance absent, tools return raw API responses without recovery hints. Response sanitization code (_strip_internal_fields) shows awareness of token efficiency, but no pagination guidance for large result sets (e.g., politician_trades, bill_votes could return hundreds of items). Missing: per-tool error classification, confirmation patterns for sensitive queries, and chaining IDs in responses.

• Document output schema for each tool. Include field names, types, and which fields are safe for chaining (e.g., bioguide_id from search_politicians → politician_profile). Use a structured format in the docstring or a separate schema reference.
• Enhance parameter descriptions with context. For min_confidence, min_tier, min_influence: explain what each score measures and why an LLM should choose one threshold over another. Reference the score-semantics docstring at the top.
• Add pagination guidance. For tools returning lists (politician_trades, bill_votes, politician_donors), document the limit parameter, explain when results are truncated, and provide a next_cursor or offset pattern if the API supports it.
• Implement error classification. Wrap API errors with actionable messages: 'Politician not found. Try search_politicians() with a partial name.' Distinguish retryable (HTTP 429, 503) from user-fixable (invalid bioguide_id) from fatal errors.
• Add tool-selection hints. In descriptions, clarify when to use search_donors vs search_donors_raw, or politician_profile vs politician_trades. E.g., 'Use search_donors for merged fingerprinted records; use search_donors_raw to see all FEC variants separately.'