Persistence for ephemeral MCP collections and records

[cdcabrera]

As a developer I need to research and review activating collection and records persistence.

Acceptance criteria

• Review, research viability of opt-in SQLite collection persistence

• If viable, create an implementation story or convert this one towards an epic

Assumptions and questions

• Opt-in for users may be the initial activation. This can include setting a DB storage path

• No model, MCP tools, MCP resources, MCP prompts, skills et all should access the DB directly. The concept should focus on being a passive store used to help in resource and server startup limitations

• Relates to

  • how the PF API has to rebuild itself on every MCP start up. Persistence would store either the crawled raw API OR be something we handled on the merged collection and records result
  • indexing collections and records for refined metadata and quality so that searches return more "exacting results"
  • moving towards 2 primary MCP resources "collections" and "records"

---

Jira Issue: PF-4590