Notebooks and LLM tools
Ways to read the Atlas from a notebook or an assistant, with the sources to point them at.
Jupyter, Colab, VS Code
Download the quickstart notebook stdlib only; the pandas cell is optional. In Colab: File, Upload notebook.
NotebookLM, ChatGPT, Claude Projects: the source pack
Add these URLs as sources (NotebookLM: Add source, Website; ChatGPT and Claude: paste or attach). They are plain text, regenerated on every publish.
| Source | What it gives the model |
|---|---|
| https://retrofitatlas.org/llms.txt | the one-page summary: what the database is, tiers, vocabularies, how to query, where to cite |
| https://retrofitatlas.org/llms-full.txt | the summary plus the full REST, MCP and data-model references |
| https://retrofitatlas.org/data/sites.csv | every record as a table (1,904 rows) |
| https://retrofitatlas.org/research/methods/ | methods, limits, known error modes |
| https://retrofitatlas.org/data/sources/ | every layer's source, vintage and licence |
| any record page | one site: before, after, timeline, sources with quotes, context layers |
Agents
Claude Code, Claude Desktop, Cursor, ChatGPT and other MCP clients connect to https://retrofitatlas.org/api/mcp; see the MCP page. AI crawlers are allowed by name in robots.txt.
Ask the model to cite
Every record has a stable id, a build date and a tier. A good prompt ends with: cite the Retrofit Atlas record URL and say whether it is curated or draft.
Documentation built 2026-10-21 from the live database (1,904 records). Corrections: the form or desk@retrofitatlas.org.