Server has 6 tools with mixed quality. Tool names follow verb_noun pattern (list_, read_, run_, write_) which is good. However, descriptions are in Chinese and vary widely in clarity. Critical gaps: most parameters lack type information in visible schemas, descriptions are often generic or domain-specific without explaining WHEN/WHY to use tools, and output schemas are not documented. Error handling is minimal. The server appears to be a research orchestrator with state-machine semantics, but this is not clearly surfaced in tool descriptions. STDIO transport caps protocol readiness at 50.

• Rewrite all tool descriptions in English with explicit WHEN/WHY guidance. E.g., 'write_stage_output: Persist structured results for the current research stage. Call this after analysis is complete and before advancing to the next stage. Must include a 'stage' field matching the current phase (ingest|fetch|calc|validate|report). Returns error if stage mismatch.'
• Document output schemas for all tools. E.g., 'list_run_files returns {files: [{name: string, type: 'json'|'md', size: number}], total: number}'.
• Add parameter constraints and examples. E.g., 'rel: Relative path within run artifacts (e.g., conflicts.json, fetch/source_1.json, calcs/01_pe.json). Must not contain .. or absolute paths.'
• Add error recovery guidance to descriptions. E.g., 'run_calculation: If function not found, check available functions via catalog. If args mismatch, review parameter names and types in function signature.'
• Implement per-tool error responses that categorize failures (retryable, user-fixable, fatal) and suggest next steps. E.g., 'stage_mismatch: Current stage is 'fetch', but write_stage_output expects 'calc'. Call list_run_files to verify current state, then retry with correct stage.'