Server has 10 tools with Polish descriptions and basic schemas, but critical gaps undermine quality. Most tool descriptions are under 50 chars (well below the 194-char baseline), providing minimal context for LLM selection. Input schemas are present but sparse, many parameters lack descriptions entirely. No output schemas documented. Tool names lack clear verb-noun structure (e.g., 'check_configuration' vs 'verify_config'). No error handling guidance, no parameter constraints (enums, ranges), and no indication of idempotency or side effects. The 'setup_api_key' and 'reset_api_key' tools expose credential handling without documented secret injection patterns. Legacy tools marked '[LEGACY]' suggest incomplete refactoring. Overall, definitions are minimal and would not guide an LLM reliably through tool selection or parameter construction.

• Expand all tool descriptions to 50 - 200 chars. Include WHAT the tool does, WHEN to use it, and any prerequisites. Example: 'check_configuration' → 'Verify Jan subagent configuration (API keys, model settings, language). Call this first to diagnose setup issues before running other tools.'
• Document output schemas for all tools. Specify returned fields, types, and structure. Example: 'write_pr_description returns {pr_id: int, description: string, status: string, created_at: ISO8601}'. This enables LLMs to chain tools and extract required data.
• Move API key handling to server-side secret injection. Remove 'api_key' parameter from setup_api_key. Instead, read NVIDIA_API_KEY from environment at startup. Document this in README.
• Add parameter descriptions for all inputs. For 'raw_notes', specify: 'Raw commit messages, PR titles, or issue descriptions to synthesize into formal documentation (1 - 5000 chars).' For 'audience', use an enum: 'Target audience: reviewer|internal|public (default: reviewer).'
• Add error handling guidance to each tool. Example: 'If API key is invalid, returns {error: "invalid_key", action: "call setup_api_key with a valid NVIDIA API key"}.' Categorize errors as retryable, user-fixable, or fatal.

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry