Critical gaps across all 18 tools. While tool names follow verb_noun convention (web, git, file, python, etc.), the schema definitions are severely incomplete. All tools declare only an 'action' enum parameter with additionalProperties:true, providing NO visibility into what parameters each action actually requires. Descriptions exist but are generic action lists ('Actions: search, scrape, read...') rather than LLM-optimized guidance on WHEN to use each tool or what it returns. No output schemas documented. No error handling guidance. No parameter descriptions beyond the action enum. This pattern suggests tools are implemented but not properly exposed via MCP, schemas are inferred from code, not formally declared.

• For each tool, replace the generic 'action' enum with explicit per-action parameter schemas. E.g., web(action='search') should declare query:string, max_results:int(1-20), include_images:bool. Use JSON Schema oneOf or discriminator patterns to express action-specific parameters.
• Document output schemas for all tools. E.g., web search returns {results: [{title, url, snippet, source}], total_count, next_cursor}. Include field types and descriptions so agents can chain calls.
• Rewrite descriptions as LLM-optimized guidance (50-200 chars). E.g., 'Search the web using Tavily AI. Use for current events, research, or fact-checking. Returns ranked results with snippets.' instead of 'Web search, scraping, and crawling operations. Actions: search, scrape, read...'
• Add error handling guidance to tool descriptions. E.g., 'If search returns no results, try broader keywords or use crawl for JS-heavy sites. If crawl times out, fall back to scrape.'
• Implement confirmation or dry-run patterns for destructive tools (cli, git write/delete, file delete, github merge, browser, workflow, schedule). E.g., add a 'dry_run' parameter or require explicit 'confirm=true' for irreversible operations.