Server has 8 tools with basic schemas and descriptions, but significant gaps prevent production use. Most tool descriptions lack actionable context (e.g., 'Query all US stock positions' doesn't explain WHEN to call vs get_tw_stocks). Parameter descriptions are sparse or missing entirely, get_trades has 4 params with minimal guidance on valid values or dependencies. No output schemas documented. Error handling returns raw exception text instead of recovery guidance. Naming is acceptable (verb_noun pattern) but lacks disambiguation for similar tools (get_us_stocks vs get_tw_stocks vs get_tw_stocks). No tool annotations (readOnlyHint/destructiveHint). add_trade is the only write tool but lacks confirmation/dry-run pattern.

• Document output schemas for all tools. Example: get_us_stocks should return {accounts: [{name, positions: [{ticker, qty, price, pl}], total_pl}], total_networth}. Include field types and whether fields are required for chaining.
• Add enum constraints to parameters. Replace 'result (Win/Loss/Breakeven)' description with inputSchema enum: ['Win', 'Loss', 'Breakeven']. Same for asset_type: ['Stock', 'Option'] and action: ['Buy', 'Sell', 'Open', 'Close', 'Roll'].
• Expand tool descriptions to 50-150 chars with WHEN/WHY context. Example: 'get_us_stocks: Returns US equity positions grouped by account (Firstrade, IBKR, etc.). Use this to review US holdings; use get_tw_stocks for Taiwan holdings; use get_networth for full portfolio overview.'
• Add tool annotations (readOnlyHint: true for all read tools; destructiveHint: true for add_trade). Helps LLMs understand safety profile without parsing descriptions.
• Implement structured error responses with recovery guidance. Example: {error: 'Invalid ticker: INVALID. Valid tickers: NVDA, AAPL, 2330.TW. Call get_quote with a valid ticker.'} instead of raw exception text.

Inferred effective spec: <=2025-11-25.

• Relies on Stateful initialize / Mcp-Session-Id (removed; protocol is stateless) - make each request self-contained