Three tools with complete input schemas and descriptions, but significant gaps in output documentation and error handling. Tool names follow verb_noun convention (extract_, ocr_, list_). Descriptions are present (90-150 chars) but lack context on when to use each tool vs. alternatives. Parameters have types and defaults, but no enum constraints for ocr_engine/schema values despite accepting only specific options. Output schemas are undocumented, LLMs cannot plan downstream calls or validate responses. No error recovery guidance. Composition is reasonable (three distinct concerns), but missing idempotency guarantees and batch variants.

• Add explicit enum constraints to 'schema' and 'ocr_engine' parameters. Replace free-form string descriptions with: schema: enum=['invoice', 'receipt', 'bill_of_lading', 'purchase_order', 'kr_tax_invoice', 'bank_statement', 'id_document', 'certificate_of_origin']. This prevents hallucinated values and makes the tool self-documenting.
• Document output schemas in tool descriptions. For extract_document, state: 'Returns a JSON object with fields matching the selected schema (e.g., invoice schema returns {vendor, date, total, line_items, ...}).' For ocr_document, state: 'Returns {text, markdown, engine, avg_confidence, block_count, table_count, processing_time_ms}.'
• Add error recovery guidance to descriptions. E.g., 'If the file is not found, verify the path. If OCR fails, try a different engine (paddle, easyocr, got). If schema validation fails, call list_schemas() to see available options.'
• Clarify when to use ocr_document vs extract_document. Add to ocr_document: 'Use this for OCR-only tasks (no LLM extraction). Use extract_document if you need structured data matching a schema.' Add to extract_document: 'Use this to extract structured data. If you only need raw text, use ocr_document instead.'
• Document idempotency. Add to extract_document: 'This tool is idempotent, calling it twice with the same file and schema returns identical results. Safe to retry on transient failures.'