Static source inference · medium confidence · detected: Logging
Deprecated protocol patterns detected
Summary
The server defines 4 tools with READ_ONLY risk classification, but critical definition gaps significantly limit production readiness. Naming is reasonable but descriptions lack LLM-focused context. Input schemas are present but lack required/type constraints at the parameter level. No output schemas are documented. Error handling and recovery guidance are absent from descriptions.
Tools (4)
consult_advisorread onlysource verified43/100
Consult a wisdom advisor based on metric, tool, or stage
Input schemas lack 'required' field constraints and do not declare parameter types explicitly (e.g., 'score' is shown as type:number but no min/max bounds). LLMs cannot validate inputs without explicit ranges.
Locate and document explicit tool registration in the SDK adapter (sdk_adapter.go or cmd/server/main.go). Show how each tool is registered with the official SDK using mcp.Tool with complete schemas.
Add 'required' constraints to all input schemas. E.g., get_wisdom should mark 'score' as required, with min=0, max=100. consult_advisor should specify which parameters (if any) are optional.
Document output schemas for each tool. Example for get_wisdom: {type: object, properties: {wisdom: {type: string, description: 'The wisdom quote'}, source: {type: string}, explanation: {type: string}}}. Include this in tool definitions.
Expand tool descriptions to 50-200 characters, explaining: WHAT it does, WHEN to use it (vs siblings), and WHAT it returns. E.g., 'Retrieve a wisdom quote and guidance tailored to your project health score. Use this for single-metric insights. For comprehensive daily guidance, use get_daily_briefing instead.'
For consult_advisor, enumerate valid metrics, tools, and stages in the parameter descriptions or use JSON Schema enum constraints. E.g., 'metric (enum): security|testing|coverage|performance'.
Add 'source' parameter to get_wisdom schema as an enum: ['pistis_sophia', 'stoic', 'random']. Document that 'random' uses date-seeded selection for consistent but varied results across multiple calls.
Add min/max bounds to numeric parameters. E.g., score: {type: number, minimum: 0, maximum: 100, description: 'Project health score from 0 (critical) to 100 (excellent)'}. days: {type: number, minimum: 1, maximum: 365}.
Spec posture evidence
Inferred effective spec: <=2025-11-25.
Relies on Logging (deprecated) - log to stderr or use OpenTelemetry
Descriptions are generic and lack context for LLM tool selection. E.g., 'Consult a wisdom advisor based on metric, tool, or stage' does not explain when to use this over get_wisdom or get_daily_briefing, or what 'wisdom advisor' returns.
Parameter 'source' in get_wisdom accepts 'random' with special date-seeded behavior. This undocumented magic behavior will confuse LLMs, it should be explicit in the parameter description and schema.
No error handling guidance in descriptions. Tool descriptions do not explain failure modes, recovery paths, or what to do if consultation fails (e.g., 'If metric unknown, call with source=random').
Parameter 'days' in get_consultation_log has no type constraint (min/max). Default is stated as 7 but not enforced in schema. Unbounded numeric inputs risk API abuse.
Parameter descriptions in 'consult_advisor' are vague. What is a valid 'metric'? What is a valid 'tool'? What is a valid 'stage'? No enum values or examples provided, LLMs will hallucinate.
consult_advisor
Add error recovery guidance to descriptions. E.g., 'If consultation fails: check that score is 0-100 and metric/tool/stage are valid. For all metrics, call with score and context only.'
Consider splitting consult_advisor into separate tools if metric, tool, and stage represent different consultation modes (consult_by_metric, consult_by_tool, consult_by_stage). If they are truly optional variants, document their interaction clearly.
Add example responses to tool descriptions or documentation to guide LLM understanding of output shape. E.g., 'Returns: {recommendation: string, confidence: number, related_tools: [string]}'.