Three tools with basic verb-noun naming (search_*, filter_*) but significant quality gaps. All tools have descriptions (10-50 chars, below the 50-200 char baseline for LLM optimization). Input schemas are present with type definitions but lack parameter descriptions entirely, a critical violation of the rubric. Output schemas are undocumented; tools return JSON strings without declaring field structure. Error handling is minimal (generic exception messages). No tool annotations (readOnlyHint, etc.). The server lacks pagination, result limits, and chaining IDs needed for agent composition.

• Add parameter descriptions to all inputs. E.g., 'job_title (string, required): The job role to search for (e.g., "Python Developer", "Data Analyst"). Partial matches are supported.' and 'location (string, default: "Hyderabad"): City or region name (e.g., "Bangalore", "Mumbai"). Accepts city names or state abbreviations.'
• Document output schemas explicitly. Declare that search_* tools return {"status": "success"|"no_results"|"error", "count": integer, "jobs": [{"title": string, "company": string, "location": string, "url": string, ...}]}. Include field descriptions.
• Add pagination: introduce 'limit' (default 20, max 100) and 'offset' (default 0) parameters to search_* tools. Return a 'total_count' field and document that results are capped at 100 per call.
• Expand tool descriptions to 100-150 chars. E.g., 'Search Naukri.com for job listings by title and location. Returns up to 20 results per call. Use filter_jobs_by_keyword to narrow results. Requires internet connectivity.'
• Add tool annotations via FastMCP decorators or schema hints. Mark search_* and filter_* with readOnlyHint=true since they do not modify state.