dartlens-mcp provides 11 tools for Korean financial disclosure (DART) queries. Tool naming follows verb_noun convention (search_company, list_disclosures, get_*). Descriptions are present but vary in quality, some are terse (e.g., 'rcept_no → 공시 본문') while others are detailed. Input schemas are visible in the code with proper JSON Schema structure (type, properties, required). However, parameter descriptions are sparse or missing in several tools. Output schemas are not documented. Error handling guidance is absent. Tool annotations (readOnlyHint/destructiveHint) are correctly implemented via LensFastMCP. The codebase shows thoughtful handling of client-side schema compatibility issues (collapsing nullable types, coercing integers to strings), but this is defensive rather than preventive, indicating upstream client bugs that the server works around.

• Document output schemas for all tools. For each tool, add a 'Returns' section describing the JSON structure, field types, and what each field means. Example: 'Returns: {company_name: str, corp_code: str (8 digits), stock_code: str (6 digits), listed: bool}'.
• Expand tool descriptions to 80-150 characters. Include WHEN to use the tool and what it returns. Example: 'Search for a company by name, stock code, or corp_code. Returns corp_code, stock_code, and company overview. Use this first to resolve company identifiers before calling other tools.'
• Add descriptions to all input parameters. For each parameter, explain the expected format, valid values, and constraints. Example: 'corp_code (str, 8 digits): The corporate code assigned by the Korean Financial Supervisory Service. Obtained from search_company().'
• Add error handling guidance to tool descriptions. Document common failure modes and recovery steps. Example: 'If company not found, try search_company() with a partial name or stock code. If disclosure not found, check the date range and disclosure type.'
• Migrate away from defensive schema transformations. File an issue with the MCP client (Claude Desktop) to fix the zod validation regression. Once fixed, remove _collapse_nullable() and coerce_integer_strings() and simplify _tool_schema.py.