Epicode MCP server has 11 tools with mixed quality. 6 tools (query_memory, cluster_detail, search_memories, cluster_similarity, check_operation, list_by_label) are defined in Rust backend with schemas visible; 5 tools (health, memory_create, memory_search, memory_recall, memory_ask) are in Python bridge with minimal schema visibility. Naming is inconsistent: some tools use verb_noun (search_memories, memory_create), others use noun_verb (query_memory, cluster_detail). Descriptions vary widely in quality, some are 10-15 chars (health, cluster_similarity), others are 50+ chars. Parameter descriptions are sparse or missing entirely. No output schemas documented. Error handling and recovery guidance absent. STDIO transport is a hard blocker for remote accessibility.

• Add output schemas for all 11 tools. Document return types, field names, and whether results are paginated. Example: search_memories should return [{memory_id, content, similarity_score, tags}] with total_count and next_cursor.
• Standardize naming to verb_noun (search_memories, query_memory → get_memory, create_memory → memory_create). Rename cluster_detail → describe_cluster, cluster_similarity → compare_clusters.
• Expand tool descriptions to 50-200 chars. Add WHEN to use each tool. Example: 'Search memories by semantic similarity. Use this to find related memories before querying specific IDs. Returns up to limit results ranked by relevance.'
• Document all parameter constraints in descriptions. Example: 'limit (integer, 1-100, default 5): Maximum results to return. Higher values increase latency.'
• Add error handling guidance. Example: 'If no memories found, try a broader query or use list_by_label to explore available tags.'