DBeast PostgreSQL MCP

What DBeast PostgreSQL MCP does

DBeast PostgreSQL MCP MCP server exposes focused PostgreSQL operations through the Model Context Protocol. It is intended for database inspection, read-only data access, query review, and operational analysis rather than unrestricted SQL execution. The server can connect to local PostgreSQL installations and hosted PostgreSQL services, including RDS or Aurora, Supabase, Neon, Railway, Render, and Fly.io when provided with a suitable connection method.

The tool set covers connection management, schema discovery, dependency mapping, query analysis, database health, security, maintenance, partitioning, data quality, replication, configuration, and audit-log access. Schema inspection can include tables, columns, indexes, relationships, and optional Mermaid ERDs. Sensitive-data scanning identifies likely PII or secrets from column names and schema patterns; it does not claim to classify data with certainty.

How it works

The DBeast PostgreSQL MCP MCP server runs locally as a Python stdio MCP server. An MCP client starts it as a subprocess, passes connection settings through environment variables or tool arguments, and receives structured results after the server queries PostgreSQL with asyncpg. No HTTP service or separate server infrastructure is required.

execute_query accepts read-only SELECT queries and automatically applies a row limit. analyze_impact examines a write query for affected rows, risk, and rollback context without executing it. Other analysis tools query PostgreSQL metadata and statistics to produce reports about performance, privileges, vacuum state, partitions, replication, configuration, and data quality.

A practical workflow is to discover the schema first, run bounded read queries, preview potentially risky changes, and then inspect health or maintenance indicators. Audit tools can retrieve logged MCP calls for a date and list available audit files.

Setup and configuration

Install the repository in editable mode with pip install -e ., then start it with dbeast or python src/server.py. Configure the MCP client with a stdio entry whose command points to the server or its source entry point. Cursor, Windsurf, VS Code, and Claude Desktop configuration examples are provided by the project.

Use DATABASE_URL for a complete PostgreSQL connection string, or provide DB_HOST, DB_PORT, DB_USER, DB_PASSWORD, DB_NAME, and optionally DB_SSLMODE. Runtime connection arguments can also supply a URL or individual connection fields. AWS Secrets Manager connections use a secret name and AWS region. The documented tuning variables include DBEAST_DEFAULT_ROW_LIMIT, DBEAST_QUERY_TIMEOUT, DBEAST_COMMAND_TIMEOUT, DBEAST_SSL_VERIFY, and DBEAST_SCHEMA_CACHE_TTL.

Tools and capabilities

• Connect, disconnect, check connectivity, inspect the connection pool, and discover local databases.
• List database structure and dependencies, including optional Mermaid relationship diagrams.
• Execute bounded read-only queries and inspect query structure, warnings, indexes, and rewrite suggestions.
• Review cache rates, connections, transaction age, slow queries, vacuum state, dead tuples, and index health.
• Inspect roles, privileges, superusers, public-schema exposure, likely sensitive columns, and PostgreSQL configuration.
• Analyze partitions, replication lag, WAL sender and receiver state, replication slots, null rates, cardinality, distributions, outliers, and duplicate rows.

Limitations and notes

The server is designed around PostgreSQL 12 or later and uses a local stdio deployment model. execute_query is limited to read-only SELECT statements, while analyze_impact previews write-query effects without performing the write. Analysis results depend on the metadata, statistics, permissions, and extensions available to the connected PostgreSQL account. The project is licensed under MIT.