Deployment

Choose the deployment shape by deciding who owns the web host.

Standalone Docker

Start from the repository .env.example, replace all sample secrets, and persist /app/data when using the default local control plane. Persist DATA_PROTECTION_KEY_PATH whenever protected identity state is enabled.

Set MCP_PUBLIC_ENDPOINT to the absolute URL MCP clients can actually reach, including /mcp. Reverse proxies may expose Admin UI, MCP, metrics, and approval callbacks through different public routes.

Standard embedded host

HsSqlAgent.Hosting embeds the same standard first-party composition and configuration contract as the official Docker image. URL binding and logging remain normal ASP.NET Core host concerns.

Use HsSqlAgent.Server instead when the application must replace authentication, authorization, UI, telemetry, middleware ordering, or the DML approval provider.

Webhook approval deployment

When DML_APPROVAL_PROVIDER=Webhook:

• the Webhook endpoint must be reachable from hs-sql-agent;
• the callback URL must be reachable by the approval service;
• the signing secret must be unique and at least 32 UTF-8 bytes;
• TLS should be used for public production traffic;
• the callback resumes protected intent and still revalidates current authorization, database configuration, policy, plan, and row-set evidence before commit.

The approval system does not receive SQL execution authority.

Scale out

For more than one application instance, use a shared Admin database and distributed coordination providers. Redis-backed cache, rate limiting, security-policy sync, outbound-delivery sync, and SQL concurrency keep process-local state from diverging across instances.

Durable pending approvals also depend on persistent Admin Store state. A restarted or different instance can resume only after current state is reloaded and evidence is revalidated.

Production checklist

• Replace HMAC_KEY, JWT_KEY, Webhook signing secrets, database passwords, SMTP credentials, and OIDC secrets.
• Persist the Admin database, data-protection keys, and audit archives where applicable.
• Configure externally reachable MCP and approval callback URLs.
• Put public endpoints behind TLS and an appropriate reverse proxy.
• Enable Prometheus and/or OTLP as required by the operating environment.
• Use distributed providers before scaling beyond one instance.

See Configuration and Distributed Deployment.