Archive MCP connector

Connect this library to ChatGPT Plus (and other MCP clients) for deeper multi-step research.

MCP (Model Context Protocol) lets ChatGPT call our archive search tools directly. You keep ChatGPT’s stronger general reasoning; the connector supplies the same historical publication index used on this site. The connector itself does not generate answers — it only returns source excerpts for the model to use.

Connector URL

Paste this URL when creating a custom MCP app / connector in ChatGPT. It is a machine endpoint, not a web page.

https://jwlib.thewt.org/mcp

Authentication: No Authentication (read-only tools). Transport: Streamable HTTP. Health check: /mcp-health.

Set up in ChatGPT Plus

Based on OpenAI’s Developer mode and MCP apps help docs. UI labels can shift slightly; use the web app (not only mobile).

Use an eligible plan. Developer mode / custom MCP connectors are available on paid plans including Plus, Pro, Business, Enterprise, and Education (web).
Enable Developer mode. In ChatGPT open Settings → Security and login and turn on Developer mode. On some accounts the toggle is under Settings → Apps → Advanced settings.
Create a custom MCP app. Go to chatgpt.com/plugins (or Settings → Apps / Connectors), click the + button, and create a developer-mode app. The + option for custom MCP apps appears only after Developer mode is on.
Enter connection details. Name: e.g. WT Library Archive. MCP server URL: https://jwlib.thewt.org/mcp. Authentication: No Authentication. Protocol: streaming HTTP / Streamable HTTP if asked.
Review discovered tools. After connect, ChatGPT should list tools such as search_articles, get_chunk, get_chunk_context, and get_article. Leave them enabled. You can refresh the app later if tools change.
Use it in a chat. Open a new conversation, choose Developer mode from the composer / tools menu, and select this app. Ask archive questions explicitly, e.g. “Use the WT Library Archive tools to find when publications first stated that Christ’s presence began in 1914 instead of 1874.”

What you get

search_articles Hybrid search over 1879–1980 publications (semantic + keyword + exact phrases). Supports year filters and oldest-first sorting.
get_chunk Fetch one indexed excerpt by chunk id, with offsets and neighbor ids.
get_chunk_context Expand text before/after a hit without loading the whole article — useful to verify quotes.
get_article Paginated plain-text fetch for one article when broader reading is needed.

Compared with the on-site AI Assistant: ChatGPT can run longer tool chains, compare more eras in one session, and apply its own reasoning style — while still grounding claims in this archive.

Other clients

Any MCP client that supports remote Streamable HTTP can use the same URL (Model Context Protocol): Cursor, Claude Desktop (with a remote MCP bridge), Codex, and similar tools.

Recommended workflow: search_articles → get_chunk_context on promising hits → dig into one article with article_id or get_article only when needed.

Notes

Tools are read-only (readOnlyHint). They cannot modify the archive.
Rate limiting applies per IP. Avoid bulk scraping; use focused queries.
Returned text is historical source material — treat it as untrusted content for the model.
If ChatGPT cannot connect, confirm Developer mode is on, the URL ends with /mcp, and health returns ok.