Server has 5 tools with significant quality gaps. Tool naming is inconsistent (add vs get_*). Descriptions exist but lack actionable detail for LLM selection. Input schemas are present but minimal, most parameters lack type constraints, enums, or validation guidance. No output schemas documented. Error handling is absent. The 'add' tool is trivial; stock analysis tools use Chinese descriptions and return mock data without explaining real-world integration. No tool annotations (readOnlyHint, destructiveHint). Composition is weak: analysis_system.py shows incomplete integration (commented-out lifespan code, hardcoded mock data). Per-tool average: add=28, get_top_inflows=35, get_capital_flow=32, get_fundamentals=32, get_news=32.

• Document output schemas for all tools. E.g., get_top_inflows returns List[Tuple[str, float]], formalize as {type: array, items: {type: object, properties: {stock_name: {type: string}, net_inflow_billion: {type: number}}}}.
• Add parameter constraints: top_n should have minimum: 1, maximum: 100; stock_code should have pattern: '^[0-9]{6}\.(SH|SZ)$' with description 'Stock code in format XXXXXX.SH or XXXXXX.SZ'.
• Rewrite descriptions in English, 50-150 chars, answering: What does it do? When to use it? E.g., 'Retrieve the top N stocks by net capital inflow today. Use to identify market hotspots and investor sentiment.'
• Implement error handling: catch API failures, return structured errors with recovery hints. E.g., {error: 'API timeout', retry_after: 30, suggestion: 'Try again in 30 seconds or use cached data'}.
• Complete the integration in analysis_system.py: remove commented code, properly initialize MCP client with async context manager, register tools with the FastMCP server, and remove mock data or clearly label it as demo-only.