Job scraper for LinkedIn, Indeed, Glassdoor, ZipRecruiter, Bayt, Naukri, and BDJobs providing job search capabilities through the Model Context Protocol
JobSpy MCP Server has well-defined tool schemas with proper parameter constraints, but lacks critical elements like output schema documentation, error handling guidance, and comprehensive parameter descriptions. Both tools are explicitly registered with JSON Schema input definitions and enums where appropriate. However, the descriptions are minimal (under the 194 char baseline for production tools), parameters lack dependency documentation, and there is no documented output structure for LLM planning. The server provides fundamental correctness (proper naming, basic constraints) but misses production-grade polish needed for reliable agent integration.
Get detailed information about a specific job posting
Search for job postings across multiple job boards
No output schema documented for tools. LLMs cannot plan downstream operations without knowing the structure and fields returned by search_jobs or get_job_details. Baseline for production tools is 100% documented return types.
Tool descriptions are too short (57 chars for search_jobs, 58 chars for get_job_details) to guide LLM selection. Baseline is 194 chars. Missing: WHEN to use, what distinguishes this from similar tools, any prerequisites, and format of returned data.
Missing descriptions on 'location' (search_jobs) and 'job_url' (get_job_details) parameters. Even though names are clear, LLMs need explicit format hints. E.g., 'location' should specify: 'City, state or just city (e.g., San Francisco, CA or Austin, Texas).' Baseline is 100% of A+ tools have param descriptions.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 56 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 43 | - | v1 |
No error handling guidance in descriptions or response examples. If no jobs are found, the tool returns 'No jobs found matching your criteria' but does not guide the LLM: should it retry with different criteria? Ask the user? This violates the recovery-guide pattern.
search_jobs lacks documented parameter dependencies. E.g., 'distance' only makes sense when 'location' is provided; 'country_indeed' only applies to Indeed/Glassdoor. These must be documented so LLMs understand constraints. Baseline from production tools: 100% document interdependencies.
No validation examples or error messages shown. If an agent passes results_wanted=101 (max is 100), or job_type='remote' (must be one of 4 enums), the tool will fail. No guidance on what error the LLM would see or how to correct it.
search_jobs description does not state that it is READ_ONLY and safe to retry. Production guidelines require describing state-modification intent so agents know retry/caching behavior.