This server has significant definition quality gaps. Only 1 of 3 tools has a well-formed input schema; the other two lack proper parameter definitions. Descriptions exist but are verbose and example-laden, violating LLM best practices. All tools lack error handling guidance, and the schema for run_code_and_show_plot is empty. Tool naming is reasonably clear, but compositions create problematic multi-step workflows. Given the tool count is small (3 tools) and 1 tool has zero input schema, the average tool score is pulled down substantially.

• Add explicit input schema validation to analyze_stock: define 'query' as a pattern-constrained string or add a pre-validation step that parses stock symbols, timeframes (1d|1w|1mo|1y), and actions (plot|analyze|compare) and rejects invalid combinations with actionable error messages like 'Invalid timeframe: got "2 weeks", expected 1d|1w|1mo|1y'.
• Replace example values in analyze_stock description with formal constraints. Instead of 'Example queries: Show me Tesla..', use: 'Stock symbols (e.g. TSLA for US, BBCA.JK for Indonesia). Timeframes: 1d, 1w, 1mo, 1y. Actions: plot, analyze, compare.'
• Add error recovery guidance: 'If stock not found, try search_stock_symbols(partial_name) or consult available tickers for your region.' This gives the agent a recovery path.
• Expand run_code_and_show_plot description to 80-150 chars: 'Executes the Python code saved by save_code() in stock_analysis.py and displays any generated plots to the user. Requires save_code() to have been called first.'
• Document output schemas for all three tools in their docstrings. Example: analyze_stock returns a string containing valid, executable Python code ready for save_code to persist. save_code returns a success message or error string. run_code_and_show_plot returns execution status and plot file path if generated.
• Add state-modification clarity to save_code description: 'Writes code to stock_analysis.py, overwriting any existing file. Side effect: modifies filesystem.'
• Implement per-tool error classification (pattern:error-classification). Instead of generic 'Error: {e}', categorize errors as: retryable (network timeout, rate limit), user-fixable (invalid symbol), or fatal (file permissions). Return structured error objects with recovery hints.