New here? Start with the Quick Start -
pip install, harborrag init, and your first answer in about ten minutes.
| You want to… | Go here |
|---|---|
| Point HarborRAG at your own sources | Configuration |
| Run ingestion and retrieval from a terminal | CLI Reference |
| Use it as a Python library | Python SDK |
| Ask questions over your content | Chat |
| Connect an IDE or agent | MCP Tools |
| Understand re-ingestion and retries | Ingestion Modes |
| Fix something that broke | Troubleshooting |
| Upgrade an existing deployment | Upgrading |
| Surface | What it does |
|---|---|
| CLI | init, doctor, chat, retrieve, and the ingest subgroup: ingest run executes in-process, the other ingest commands drive Temporal |
| HTTP API | /v1/... public contract - ingestion, retrieval, chat, agent, admin; /api/v1/... operational routes |
| MCP | nine read-only evidence, source, and graph tools over stdio or bearer-authenticated loopback HTTP, plus a browser playground |
The MCP server implements stdio plus bearer-authenticated, loopback-only Streamable HTTP. TLS, remote exposure, and production token verification remain deployment-owned. The project status records the current boundary.
For direct Python use of connectors, parsers, model clients, and repositories, read
packages/harborrag-adapters/README.md and
the family README nearest the implementation. To add a new provider, see
Extending HarborRAG.