Dbridge MCP

What ugurcl/dbridge-mcp MCP server does

ugurcl/dbridge-mcp MCP server gives an MCP-compatible client a guarded, read-only view of a SQL database. It supports SQLite, PostgreSQL, and MySQL/MariaDB, using the connection argument to determine which engine to use. A file path selects SQLite, while postgres://, postgresql://, and mysql:// connection strings select the corresponding network database.

The server is suited to natural-language data questions because the agent can first inspect the available schema and then produce a query for the user’s request. It also exposes database metadata and performance information for tasks that go beyond retrieving rows.

How it works

Schema discovery is available through table listing, table descriptions, and a schema resource. Query execution accepts one SELECT or WITH statement and can format results as JSON, CSV, or Markdown. Query plans can be inspected without executing the query, while row counts, samples, column statistics, index information, and recorded slow statements support analysis and troubleshooting.

ugurcl/dbridge-mcp MCP server applies multiple restrictions around database access. The SQL guard rejects writes, DDL, and data-modifying common table expressions. Queries use read-only database access, and PostgreSQL and MySQL queries also run in read-only transactions. Row limits still apply when a submitted query contains a larger LIMIT. PostgreSQL and MySQL queries have per-query timeouts, and expensive queries can be rejected based on estimated cost.

Column privacy can be configured by hiding columns from the model or masking their values while keeping them available for querying. Rate limits and a capped connection pool further limit request volume and concurrent database use.

Setup and configuration

The package requires Node.js 22.5 or newer. SQLite uses Node’s built-in node:sqlite, so the documented npm runner does not require a native compilation step. The published package can be started with npx -y dbridge-mcp followed by a database path or connection string.

A configuration file can be supplied through the DBRIDGE_CONFIG environment variable. The README shows this variable in Claude Desktop, Claude Code, and OpenCode configurations. MCP clients start the process as a subprocess; examples are provided for Claude Desktop, Cursor, Windsurf, and Claude Code. Docker usage is also documented, including read-only mounting of a configuration file.

Tools and capabilities

The available tools include:

• list_tables, describe_table, sample_table, and count_rows for schema and table inspection.
• run_query for one read-only query with selectable output formatting.
• explain_query for a plan and estimated cost without execution.
• column_stats for distinct-value counts and null fractions.
• index_health for index sizes, scan counts, and unused, duplicate, or invalid index flags.
• test_index for PostgreSQL hypothetical index testing through HypoPG.
• slow_queries for recorded statement timings and call counts where PostgreSQL pg_stat_statements or MySQL performance_schema is available.
• get_limits for reporting active caps, timeouts, and column privacy settings.

The server also provides the dbridge://schema resource and an optimize prompt for a guided performance review. The prompt examines slow queries, plans, and index health, then validates index recommendations with test_index before suggesting them.

ugurcl/dbridge-mcp MCP server is a fit for read-only analytics, private-data question answering, and database diagnostics. It is not intended for migrations, updates, inserts, deletes, schema changes, or other write workflows.