Server has 4 tools with complete input schemas and descriptions. Naming follows verb_noun convention (search_jobs, get_job_details, match_candidate_cv, get_salary_benchmarks). Descriptions are detailed (100-250 chars) and explain context well. However, output schemas are not documented, responses are inferred from code inspection. Parameters lack enum constraints where applicable (category, district, experienceLevel). Error handling is minimal; no recovery guidance or actionable error messages visible. Tool composition is reasonable but match_candidate_cv lacks clarity on what 'match score' means operationally.

• Add explicit output schema documentation for all 4 tools. Example: 'Returns array of {id: string, title: string, company: string, district: string, salary: {min: number, max: number}, url: string, source: "live_zangia" | "local_mcp_db"}'.
• Convert category parameter to enum: ['IT_TELECOM', 'BANKING_FINANCE', 'MINING_HEAVY_INDUSTRY', ...]. Same for district: ['Sukhbaatar', 'Chingeltei', 'Khan-Uul', 'Unknown', 'Remote']. Enums are self-documenting and prevent hallucination.
• Clarify match_candidate_cv output: does it return {matchScore: number, skillGaps: [{skill: string, gap: string}]} or a different structure? Add example output to description.
• Add error handling guidance: 'If job ID not found, returns {error: "Job not found"}. Try search_jobs() to discover valid IDs.' Same for CV validation and empty benchmark results.
• Refactor search_jobs minSalary logic: either exclude live jobs from salary filtering (since they have salary=0), or document that minSalary only applies to local_mcp_db jobs. Current behavior is ambiguous.