# nanograph-mcp

Build Nanograph datasources from Markdown or PDF corpora and serve them through
MCP or a small HTTP/JSON wrapper.

## What v1 supports

- Ingest commands: `nanograph-ingest seed --config ...`, `build`, and `smoke`
- Server commands: `nanograph-mcp --registry ...` and `nanograph-http --registry ...`
- MCP tools: `sources`, `source`, `documents`, `document`, `sections`, `search`, `passage`
- HTTP routes: `GET /healthz`, `GET /datasources`, `GET /datasources/{name}`,
  `GET /datasources/{name}/documents`,
  `GET /datasources/{name}/documents/{document_id}`,
  `GET /datasources/{name}/documents/{document_id}/sections`, `POST /search`,
  `POST /passage`, and `POST /mcp`

Use `/healthz` for health checks. Other application responses use the same
schema-versioned envelope.

Search modes:

- `mode = "keyword"` for all datasources
- `mode = "search"` only for datasources built with embeddings enabled

## Quick start

```sh
nanograph-ingest smoke --config examples/corpus-markdown.toml
nanograph-http --registry examples/datasources-public.toml --host 127.0.0.1 --port 8765
```

The public registry example uses only `home`, `exposure`, and `enabled` for
each datasource. Host-local path overrides such as `database` and `config` are
outside the documented v1 contract.

Documentation:

- [API reference](docs/api.md)
- [Deployment notes](docs/deployment.md)