LabscoMCP early access ↗
iPraBhuMCP SERVER · OFFICIAL PROJECT

MCP Database Server by iPraBhu

One MCP server across PostgreSQL, MySQL, SQLite, SQL Server and Oracle — cached schema discovery, join suggestions, and read-only by default.

Relational SQL Databases✓ Official

LABSCO SUMMARY

Three layers of write protection, and you have to defeat all of them.

That is the right shape for a tool an assistant drives: read-only per database, writes off globally, and the destructive statements still blocked after you enable writes. The schema cache is the other design decision that shows up in use — introspection is expensive on a large database, so caching it with a TTL and a version hash means the join suggestions and query help are answered from a local copy rather than re-read every turn. If you are wiring one database only, this is more configuration than you need; the payoff starts at two.

What it is

A SQL database MCP server built for the case where there is more than one database. Each connection gets an ID in a config file, and every tool call names which one it means. It discovers and caches schemas, infers relationships from foreign keys and heuristics, and blocks writes unless you have explicitly turned them on.

What you get

  • The configured databases listed with their connection and cache state, and schema discovery that caches with a TTL and a version hash — list_databases, introspect_schema, get_schema, cache_status, clear_cache
  • Parameterised queries with row limits, offsets, a byte cap on the response and a per-call timeout, plus a streaming export of large read-only results to JSONL or CSV — run_query, export_query
  • Join paths suggested from the relationship graph for a set of tables you name, rather than worked out from the schema by hand — suggest_joins
  • Query analysis: execution plans, index recommendations, slow-query detection, rewrite suggestions and per-query profiling — explain_query, suggest_indexes, detect_slow_queries, rewrite_query, profile_query, analyze_performance
  • A connectivity test for when the failure is the connection rather than the query — health_check

Requirements

A .mcp-database-server.config file describing each database — an id, a type of postgres, mysql, sqlite, mssql or oracle, and a connection string. Credentials have three paths and only one is recommended: secretRef names an environment variable, credentialCommand runs a shell command that prints the string (1Password, Vault, an AWS helper), and an inline url is supported but not the recommended secret handling. Install with npm install -g @adevguide/mcp-database-server or run through npx; the package is @adevguide/mcp-database-server (2.0.2) with a mcp-database-server binary. Writes are off by default at two levels — per-database readOnly defaults to true and the global allowWrite defaults to false — and disableDangerousOperations still blocks DELETE, TRUNCATE and DROP even when writes are allowed. Oracle is a stub requiring Oracle Instant Client.

Setup effort

One command plus credentials — npm install -g @adevguide/mcp-database-server, then supply credentials

See installation options