SchemaCrawler

Connects to relational databases (PostgreSQL, MySQL, Oracle, SQL Server, etc.) through the Model Context Protocol and introspects complete schema metadata including tables, columns, constraints, indexes, and relationships. Uses JDBC drivers to query system catalogs and information schemas, then serializes schema objects into structured JSON/text representations that LLM agents can reason about and query. Enables AI systems to understand database structure without manual schema documentation.

Generates syntactically and semantically valid SQL queries by providing the LLM with complete schema context including column types, constraints, and relationships. The MCP server exposes schema metadata that the LLM uses to construct queries that respect database structure, avoiding common errors like invalid column references, type mismatches, or constraint violations. Works by embedding schema information in the LLM's context window so it can generate queries that match the actual database structure.

Enables natural language questions about database schema semantics and metadata, such as 'what does the USR_PREFIX column mean?' or 'which tables store customer information?'. The MCP server provides schema metadata to the LLM, which uses its reasoning capabilities to answer questions by analyzing column names, types, relationships, and any available documentation or comments. Works by exposing schema objects as queryable resources that the LLM can search and reason about.

Implements the Model Context Protocol (MCP) server specification to expose database schema as queryable resources that MCP-compatible clients (Claude Desktop, custom agents, etc.) can discover and interact with. Uses MCP's resource and tool abstractions to represent tables, columns, and relationships as first-class entities with defined schemas and capabilities. Enables seamless integration between LLM applications and databases through a standardized protocol.

Manages connections to multiple relational databases simultaneously through a single MCP server instance, supporting different database systems (PostgreSQL, MySQL, Oracle, SQL Server, etc.) with database-specific JDBC drivers. Routes schema introspection and query requests to the appropriate database based on connection configuration. Enables agents to work with heterogeneous database environments without separate server instances.

Provides configurable filtering and scoping of schema introspection results to focus on relevant tables, columns, and schemas based on patterns, inclusion/exclusion rules, or explicit selection. Uses regex or glob patterns to match schema objects and reduce the amount of metadata exposed to the LLM, improving context efficiency and reducing noise. Enables agents to work with large databases by focusing on specific subsets.

Caches introspected schema metadata in memory to avoid repeated expensive database queries, with configurable refresh intervals or manual refresh triggers. Enables fast responses to repeated schema queries while maintaining freshness through periodic or event-driven updates. Balances performance with accuracy for long-running agent sessions.

Analyzes column naming patterns and prefixes (e.g., USR_, ORD_, CUST_) to infer semantic meaning and categorize columns by business domain. Uses pattern recognition and naming convention analysis to help LLMs understand what column prefixes represent without explicit documentation. Enables semantic reasoning about column purposes based on naming conventions.

Exposes foreign key relationships, primary keys, unique constraints, and check constraints as queryable metadata that LLMs can use to understand data relationships and generate valid queries. Represents relationships as graph-like structures that agents can traverse to understand data dependencies and cardinality. Enables semantic reasoning about data integrity and referential relationships.

Exposes index definitions, column statistics, and performance-related metadata to help LLMs understand query optimization opportunities and avoid inefficient query patterns. Provides information about indexed columns, index types (B-tree, hash, etc.), and cardinality statistics that agents can use to reason about query performance. Enables AI systems to generate more efficient queries.