Overview
This page is the system-level reference: how data flows during conversion, how an agent query is served, and how a project transitions through its lifecycle.
High-level flow
The conversion (write) path and the agent-query (read) path share one database.
- The app calls
POST /api/convert, which creates a project + token and kicks off a background pipeline (crawl → chunk → analyze) writing to Postgres. - Agents call
GET/POST /api/mcp/{projectId}/mcpwith the Bearer token; the endpoint reads only that project's data.
Request sequence
A single tools/call from an agent flows through the MCP endpoint and retrieval
layer and back as a cited result.
Conversion lifecycle
The status field drives the convert page UI and is persisted on every
transition, so polling or resuming just reads the row.
