Server has 11 tools with mostly complete schemas and descriptions. Naming follows verb_noun convention (ib_*). Descriptions are present and substantive (avg ~120 chars). Input schemas use Pydantic models with type constraints and enums. However, output schemas are not documented, LLMs cannot predict response structure. Error handling is minimal; no recovery guidance. Two-step order flow (prepare/submit) is well-designed for safety but lacks explicit confirmation prompts in tool descriptions. Missing tool annotations (readOnlyHint, destructiveHint) despite clear risk levels. Parameter descriptions are good but some lack format/range details (e.g., expiry format YYYYMMDD stated but not enforced in schema).

• Document output schemas for all tools. Add a 'Returns' section to each tool docstring specifying field names, types, and meaning. E.g., ib_account_summary returns {tag: str, value: float, currency: str}[].
• Add tool annotations via FastMCP. Use @mcp.tool(readOnlyHint=True) for read tools, destructiveHint=True for cancel/submit, idempotentHint=True for status checks. This enables LLM safety reasoning.
• Wrap all tool implementations in try-except blocks that return structured error objects with 'error_code', 'message', and 'recovery_hint'. E.g., 'Connection failed. Ensure TWS is running on 127.0.0.1:7496. Retry in 5 seconds.'
• Add Pydantic validators and enums for constrained fields. Replace bar_size string with Enum(1_SECS, 5_SECS, 1_MIN, ...). Add regex pattern for expiry: Field(..., pattern=r'^\d{8}$'). This prevents hallucinated invalid values.
• Enhance ib_submit_order to validate token expiry and match against pending order details. Return clear error if token is expired or does not match the order being submitted.