Server exposes 6 read-only tools for Riposte language discovery via STDIO. Tool names follow verb_noun convention (language_overview, list_topics, get_topic, predicate_reference, explain_error, check_program). Descriptions are present and contextual (avg ~80 chars), but parameter descriptions are sparse or missing. Input schemas are minimal, most tools accept 0-1 parameters with basic type info. Output schemas are not documented. No error handling guidance or recovery paths. Tools are well-scoped (single responsibility) and compose logically (list_topics → get_topic chain). Security is adequate for read-only operations, but no input validation or bounds checking visible.

• Add parameter descriptions to get_topic, predicate_reference, and explain_error. Example: 'name: The steering topic name (e.g., "types_and_estimates", "quirks"). Call list_topics() to see available options.'
• Document output schemas for all tools. Example: 'Returns: {"topics": [string]}, a sorted list of topic names.' or 'Returns: string, the markdown content of the overview.'
• Add input validation to check_program: enforce max source length (e.g., 100KB), set a timeout (e.g., 10s), and return structured error responses with line/column info and recovery hints.
• Enhance error messages in explain_error to include common fixes or related predicates. Example: 'E030: Type mismatch. Expected tribool (likely/worst_case/guaranteed), got boolean. Did you mean: likely(can_ko(...)) instead of can_ko(...)?'
• Consider HTTP+SSE or Streamable HTTP transport for remote accessibility. STDIO limits integration with hosted agents and cloud platforms.