Three tools with solid naming (verb_noun pattern) and comprehensive descriptions (150-250 chars each). All tools have input schemas with type definitions and parameter descriptions. However, output schemas are not explicitly documented in the source code, only inferred from implementation. Parameter constraints (e.g., debt_window_years range 2-10) are documented in descriptions but not enforced in JSON Schema. Error handling is minimal; tools return {"error": "..."} dicts but lack recovery guidance or categorization. No tool annotations (readOnlyHint, etc.) despite all three being read-only operations.

• Document output schemas for all three tools. For get_fundamentals, specify that it returns {ticker, company_name, pe10, pfcf10, peg, debt_to_equity, ...} with types and descriptions. Use JSON Schema definitions or explicit prose.
• Add readOnlyHint: true to all three tool schemas to signal they are safe for agents to call without confirmation.
• Enhance error responses with recovery guidance. E.g., 'Ticker INVALID not found. Call list_available_indicators or search for similar tickers.' Categorize errors: retryable (network timeout), user-fixable (invalid ticker), fatal (auth failure).
• Add minimum/maximum constraints to JSON Schema for debt_window_years (minimum: 2, maximum: 10) instead of relying on description text alone.
• For screen_companies filters, define an enum or additionalProperties schema that constrains filter keys to the valid set (pe1, pe2, ..., debt_to_equity, etc.) so LLMs can discover options without a separate call.