MCP Phish exposes only one tool: 'health'. The tool name lacks a clear action verb (should be 'check_health' or 'get_health_status'). The description is verbose (180+ chars) but does not follow the pattern:tool-description guidance of 10 - 1024 chars optimized for LLM selection. No input schema is visible in the provided source code, the tool appears to accept no parameters, but this is not explicitly documented. No output schema is documented. The server is HTTP-based (fastmcp), which is positive for transport, but the single tool's definition quality is severely limited by missing schema documentation and unclear naming.

• Rename 'health' to 'check_server_health' or 'get_health_status' to start with a clear action verb (pattern:tool).
• Add explicit JSON Schema for the output: document fields like {status: string, cache_hit_timestamp: string, cache_miss_timestamp: string, upstream_reachable: boolean, vault_staleness_seconds: number}.
• Shorten description to ~80 - 120 chars: 'Check server health status including cache performance, upstream API reachability, and vault staleness.' Remove Docker/implementation details.
• If the tool accepts no parameters, explicitly document that in the schema (empty properties object) so LLMs know not to pass arguments.
• Add error handling guidance: document what status values are possible (e.g., 'healthy', 'degraded', 'unhealthy') and what each means for the agent's next action.

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

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