hs-sql-agent is an HTTP MCP server for governed SQL access. It connects MCP clients to PostgreSQL, MySQL, SQL Server, Oracle, SQLite, and Firebird while keeping identity, access policy, SQL validation, mutation approval, and database execution on the server side.
Generated SQL is untrusted input. The model may propose SQL, but the server decides whether the request is authorized, semantically supported, and safe to execute.
Three ways to use the product
Deploy standalone
Use the official Docker image or repository Compose setup when hs-sql-agent should run as its own service. This is the shortest path to the complete first-party product.
Embed the complete product
Use HsSqlAgent.Hosting when a .NET application should embed the same standard first-party composition and configuration contract as the official Docker image.
Build a custom integration
Use HsSqlAgent.Server when an existing ASP.NET Core application must own authentication, authorization, middleware ordering, frontend, telemetry, or approval-provider composition. Start with AddHsSqlAgentCore() and add only the capabilities the host needs.
Security boundary
Queries go through parsing, binding, authorization, policy validation, capability proof, provider-specific compilation, and execution. Unsupported syntax or semantics fail closed instead of falling back to raw provider execution.
DML is a separate risk class. execute_dml_sql can accept one or more supported semicolon-separated mutations, but the whole request still goes through server-owned approval evidence, revalidation, and atomic transaction execution.
Approval adapters never receive a database connection, transaction, validated execution plan, or commit primitive. hs-sql-agent retains ownership of the execution boundary.