Two tools with clear, well-structured definitions and good descriptions. Both tools have complete input schemas with typed parameters and descriptions. However, output schemas are not explicitly documented, and error handling guidance is minimal. Tool naming follows verb_noun convention (parse_document, extract_information) appropriately. Descriptions are substantive (185-280 chars) and explain what each tool does, when to use it, and supported formats. Parameters are well-described with type information. Key gap: no documented output schema structure, no error recovery guidance, and no validation constraints on inputs (e.g., file format validation, file size limits mentioned in description but not enforced with schema constraints).

• Document the output schema for parse_document: specify that it returns a structured object with fields like 'content' (string), 'metadata' (object with document type, page count, etc.), and 'format' (the requested output format). Use JSON Schema in a separate 'output' field if the MCP framework supports it.
• Document the output schema for extract_information: clarify whether it returns a JSON object matching the provided/generated schema, or a wrapper object with { 'extracted_data': {...}, 'schema_used': {...} }. Specify field types.
• Add JSON Schema constraints to file_path parameters: use pattern validation (e.g., regex for supported extensions) or enum the allowed formats. This prevents LLMs from attempting to process unsupported files.
• Add maxLength constraint to file_path to prevent path traversal attacks: e.g., maxLength 512.
• In extract_information, explicitly document the precedence for schema inputs: 'If schema_path is provided, it takes precedence over schema_json. If both are omitted and auto_generate_schema=true, a schema will be generated automatically. If both are omitted and auto_generate_schema=false, the tool will return an error.'
• Add error recovery guidance in both tool descriptions. Example: 'If the API returns a 401 error, verify UPSTAGE_API_KEY is set. If the file format is unsupported, try converting to PDF first. If the file is too large, split it and process each part separately.'
• Consider splitting extract_information into two tools: extract_with_schema (requires schema_path or schema_json) and extract_auto (auto-generates schema). This reduces parameter ambiguity and makes tool intent clearer.