Skip to content
hs-sql-agent

MCP-native · fail-closed · six SQL dialects

Secure SQL execution for AI agents.

Put a compiler and governance boundary between generated SQL and your databases. hs-sql-agent parses, validates, proves capabilities, applies access policy, and only then executes provider-specific SQL.

docker compose up -d

Compiler boundary

Treat generated SQL as untrusted input.

The important part is not generating SQL. It is proving that the statement is inside the dialect, capability, and policy boundary you intended to expose.

Safe mutation path

DML approval is a protocol, not a confirmation dialog.

UPDATE and DELETE are previewed without mutation, approval is bound to the validated compiled plan and row-set fingerprint, and the matched rows are checked again before commit.

01
Preview

Read the affected rows without executing the mutation.

02
Approve

Require explicit MCP form Elicitation from the human operator.

03
Revalidate

Re-query inside the commit transaction and compare the bound row set.

04
Commit

Execute only when the plan, policy, challenge, row count, and row set still match.

One surface, six dialects

Use the database you already run.

The compiler keeps provider-specific SQL semantics explicit while MCP clients interact through a single governed interface.

MCP and .NET integrations

Meet agents where they already work.

Connect remote MCP clients such as Claude Desktop and Cursor, or embed the same governed server boundary inside an ASP.NET Core application.

Documentation

From first MCP key to production governance.

The documentation is organized around the way operators actually adopt hs-sql-agent: setup, MCP, compiler guarantees, administration, integrations, operations, and development.