Harvest MCP exhibits significant definition quality gaps. Tool descriptions are present but generic (avg ~80 chars, below 194-char baseline). Input schemas are visible but lack parameter descriptions, critical for LLM reasoning. No output schemas documented. Tool names follow verb_noun convention (good), but parameter naming is inconsistent (e.g., 'extraction' enum vs 'format' would be clearer). Error handling and recovery guidance absent. No tool annotations (readOnlyHint/destructiveHint). The 'denseforge_ingest' and 'schedule' tools are WRITE operations but lack confirmation/dry-run patterns. Batch tool (harvest_batch) lacks pagination guidance. Overall, definitions are skeletal, sufficient for basic invocation but insufficient for robust agent planning.

• Add output schema documentation for all tools. Example: harvest_scrape returns {content: string, url: string, title?: string, metadata?: object}. This enables agents to chain tools and extract required fields.
• Expand parameter descriptions to 50 - 150 chars with format/constraint guidance. E.g., 'selector: CSS selector string (e.g., "div.product"). If no match, tool returns empty content. Use harvest_scrape first to inspect page structure.'
• Implement confirmation/dry-run for WRITE tools. denseforge_ingest should accept a dry_run=true flag; schedule should return a preview of the cron schedule before confirming.
• Add error recovery hints to tool descriptions. E.g., 'If extraction fails, try harvest_scrape with a narrower selector, or use harvest_extract with a more specific prompt.'
• Declare tool annotations (readOnlyHint, destructiveHint, idempotentHint) in tool definitions. Mark denseforge_ingest and schedule as destructiveHint=true; mark harvest_scrape and proxy_list as readOnlyHint=true.

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

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry