Taruvi backend provisioning

Use the Taruvi MCP server to provision and modify backend resources. This skill covers tool selection, invocation order, the shapes the tools expect, and the gotchas that don't come through in tool descriptions.

This skill is the control plane. If you're writing Python that runs inside a deployed function, switch to taruvi-functions. If you're building a Refine frontend, switch to taruvi-refine-frontend.

Compliance rule: This skill's prescribed patterns (Frictionless schemas, MCP tool invocation order, destructive-op protocol) are mandatory. Do not invent endpoints, skip validation, or hardcode secrets. If a requirement cannot be met, stop and ask the user.

Core principles

1. Trust the MCP tool, not your memory. The MCP server is authoritative for Taruvi's current behavior. Read tool responses carefully — they return structured IDs, slugs, and status you need for the next step.
2. Verify before destructive ops. delete_datatable, schema changes that drop columns, policy overwrites, and raw SQL DDL all cause irreversible or hard-to-reverse changes. Always plan → validate → execute for these.
3. Frictionless schemas are upserts, not merges. create_update_schema replaces table definitions — fields missing from your new payload are dropped, not preserved. Always inspect the current schema first.
4. Cerbos policies are REPLACEMENTS. manage_policies(action="create_update") fully replaces the policy body, not merged.
5. Tenant context is ambient. The MCP server resolves tenant + app from the session context. Do not try to pass tenant slugs in tool arguments unless a specific tool accepts app_slug.

Tool index (24 tools)

Grouped by domain. See references/mcp-tool-quickref.md for one-line signatures.