Multi-database MCP server for PostgreSQL, MySQL, MongoDB, MariaDB, SQLite, and Redis with per-connection access modes and performance analysis tools
39 tools with consistent naming (db_* prefix) and basic descriptions, but critical gaps in schema documentation and parameter descriptions. All tools have input schemas visible in tool_names.py, but descriptions are generic (avg ~60 chars, well below 194-char baseline). Most parameters lack detailed descriptions, e.g., 'conn' is described only as 'Connection name' with no guidance on valid values or format. No output schemas documented. Error handling and recovery guidance absent. Composition is reasonable (single responsibility per tool), but parameter constraints (enums, ranges, formats) are missing. Security concerns: no evidence of secret injection patterns or permission gates for destructive tools (db_admin_kill, db_admin_cluster, db_write_query).
Run ANALYZE or equivalent to update table statistics
Perform cluster administration operations (rebalance, failover, etc.)
Kill a running query or connection
Run maintenance operations (VACUUM, OPTIMIZE, etc.)
List currently active operations and queries
Identify blocking queries and locks
Get cluster topology and node status (for clustered databases)
Parameter descriptions are minimal (avg ~20 chars). 'conn' appears in 39 tools with only 'Connection name', no guidance on valid values, format, or how to discover available connections. LLMs cannot infer whether to pass 'postgres_prod', 'pg_app', or a UUID.
No output schemas documented. Tools return results but LLMs cannot plan downstream calls or extract required fields. E.g., db_schema_connections returns a list but structure is unknown, does it include connection_id, uri, status, or just name?
Destructive tools (db_admin_kill, db_admin_cluster, db_write_query) lack permission gates, dry-run support, or confirmation patterns. No evidence of access control or audit logging. An agent could kill production queries without authorization checks.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 49 | 2026-07-28+ | v2 |
Get column-level statistics (cardinality, distribution, etc.)
Validate database configuration against best practices
Run diagnostic checks on database health and configuration
Explain query execution plan and performance
Check overall database health status
Get recommendations for index creation and optimization
Get index usage statistics and recommendations
Retrieve database logs and error messages
Check replication status and lag (for replicated databases)
Get database configuration settings
Get storage usage and allocation information
Get table or collection statistics (row count, size, etc.)
Check TLS/SSL certificate and encryption status
List top queries by execution time or frequency
Get statistics on vector indexes and embeddings
Export query results as CSV or JSON
Execute a read-only query (SELECT for SQL, find for MongoDB)
Sample random rows from a table or collection
Perform full-text search (MongoDB $search, MySQL FULLTEXT, PostgreSQL FTS)
Perform vector similarity search (MongoDB $vectorSearch, PostgreSQL pgvector)
List configured database connections
List databases in the connected server
Get the DDL (Data Definition Language) for a table or collection
Describe the schema of a table or collection
List user permissions and grants
List tables, collections, or other objects in a database
List foreign key relationships and constraints
Search for tables, columns, or collections by name or pattern
List search indexes (MongoDB Atlas $search, $vectorSearch)
List database users and roles
Execute a write query (INSERT, UPDATE, DELETE for SQL; insert, update, delete for MongoDB)
Create or update search indexes (MongoDB $search, $vectorSearch)
No error handling guidance. Tools lack recovery hints (e.g., 'Connection not found. Call db_schema_connections to list available connections.'). LLMs cannot self-correct on failures.
Parameters lack constraints (enums, ranges, formats). 'limit' appears in 5+ tools with no min/max bounds, LLMs could pass limit=999999 causing timeouts. 'format' in db_read_export has no enum constraint.