Atlas is a public corpus of editorial research drafts. Every article decomposes into atomic claims with evidence locators and the exact source versions they came from, allowing your coding agent to cite primary evidence instead of paraphrasing from training memory.

> **Drafts, not consensus:** Atlas articles are editorial research drafts. Source inspection and claim standing describe the recorded evidence; they do not establish independent replication or clinical/scientific consensus.

## Query from your coding agent (MCP) [#query-from-your-coding-agent-mcp]

Once connected via the [quickstart](/starlight/quickstart), your agent has `atlas_search` and `atlas_get_article`. Ask:

```text
Search Starlight Atlas for memory, open the most relevant article, and list its atomic claims with their source revisions.
```

## Query from the HTTP API [#query-from-the-http-api]

```bash
curl "https://starlightintelligence.ai/api/v1/atlas?q=memory&limit=5"
```

| Parameter | Constraint & meaning |
|---|---|
| `q` | Keyword query, up to 200 characters |
| `domain` | Filter to a specific research domain (up to 200 characters) |
| `lens` | One of `foundation`, `frontier`, `historical`, `hypothesis` |
| `limit` | Page size from `1` to `50` |
| `offset` | Pagination offset from `0` to `10000` |

## Bulk corpus & bibliography exports [#bulk-corpus--bibliography-exports]

- **Public JSON corpus:** `https://starlightintelligence.ai/atlas/corpus.json` — summaries and bibliographic records with source-specific rights preserved.
- **CSL JSON citations:** `https://starlightintelligence.ai/atlas/citations.csl.json` — import directly into Zotero, Pandoc, or any Citation Style Language tool.
- **Methodology:** `https://starlightintelligence.ai/atlas/method` — how sources are inspected, version-pinned, and reviewed.