Database Integration

What you'll accomplish: Understand how (and whether) Hoziron agents reach a database — and why there's no built-in "database connector" type to configure.

There is no first-class database connector

Unlike some platforms, Hoziron has no dedicated "database integration" object, no postgresql-connector/mysql-connector built-in package, and no [database] config section. Searching the kernel and platform code for a database-specific connector abstraction turns up nothing — the only integration surface that exists is the generic one described in APIs: an integration-type package that exposes named tools, either via an MCP server (payload/mcp.json) or a REST/SOAP adapter contract (payload/adapter.json).

If a postgresql-connector-style package exists for your deployment, it's because someone (Hoziron, your team, or a vendor) authored and published one as an ordinary integration package — not because the platform ships one. Check hoziron-cli catalog search "postgres" (or your organization's private registry) rather than assuming a name.

The real mechanism

If your workflow needs database access, the model is:

  1. Author or install an MCP server package that talks to your database and exposes typed tools for the specific operations your agents need (e.g. claims_db.search_claims, claims_db.get_policy_row) — not a raw, unbounded execute_sql tool. This is the same shape as the reference claims-core/policy-admin MCP contracts described in APIs § MCP servers; a database-backed integration is architecturally identical to those, just with a SQL client instead of a carrier API client behind the tool implementation.
  2. Declare the contract or integration on the consuming competency (contracts: if the backing database may vary by deployment, integrations: if it's fixed).
  3. Store the connection string/credentials in the vault, referenced by name — never inline. See Credentials.
  4. Install and connect it like any other integration:
hoziron-cli catalog install claims-db-mcp
hoziron-cli vault set CLAIMS_DB_CONNECTION_URL
hoziron-cli integration connect claims-db-mcp
hoziron-cli integration tools claims-db-mcp

Why tool-shaped, not query-shaped

Every outbound call — including a database read — passes through the mediation seam, where PII is detected and tokenised by default before it reaches a model, and hydrated only for named destinations with an explicit hydrate rule (PII Engine). That seam operates on the tool call's arguments and response, not on parsed SQL — so a well-designed database-backed integration exposes narrow, typed tools (get_claim_by_id, search_open_claims) rather than a generic query executor. This also keeps the least-privilege story concrete: a competency's allowed_tools boundary can name claims_db.search_claims specifically, something it cannot do against an arbitrary SQL string.

Regardless of how the database is reached, use a read-only database role for any integration that doesn't need to write:

CREATE ROLE hoziron_reader LOGIN PASSWORD '...' IN ROLE pg_read_all_data;

hoziron-cli vault set CLAIMS_DB_CONNECTION_URL
# Enter: postgresql://hoziron_reader:...@db.internal:5432/claims

Next steps

Related: