Scored across 191 tools
Many tools share similar purposes: multiple PG19 feature status probes (get_pgq_status, get_repack_status, get_aio_status, get_pg19_stats_status, etc.), several query analysis tools (explain_query, analyze_query_plan, why_is_this_slow, optimize_query), and numerous vector search variants. While descriptions are detailed, the sheer number of similarly named tools creates real ambiguity for an agent selecting the right one.
Names are consistently snake_case, but the convention is mixed: many tools use verb_noun (list_tables, run_select, generate_prisma_schema), while others are noun-first (vector_search, hybrid_search, full_text_search, mmr_search) or sentence-like (why_is_this_slow). Still readable, but not a predictable pattern.
191 tools is far beyond any reasonable scope for a single MCP server; even a production-grade PostgreSQL server should consolidate overlapping capabilities. This extreme count overwhelms agents and greatly increases misselection risk.
The surface covers an extensive range of PostgreSQL features: introspection, performance advisors, vector/RAG, extensions, PG19 features, graphs, time-series, and more. Minor gaps may exist (e.g., direct DDL for roles/databases), but overall coverage is very broad and leaves few dead ends.
