Time Machine News
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@Time Machine News What did newspapers report the week the Titanic sank?"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
time-machine-news
MCP server that lets Claude explore historical US newspapers from the Library of Congress.
It connects Claude to Chronicling America, the Library of Congress archive of more than 20 million digitized newspaper pages from the 1700s to 1963. Ask about a day, a town, or an event, and Claude searches the archive, reads the pages, and links back to each scan.
Use the hosted version
Open https://time-machine-news-alpha.vercel.app and copy the connector link, or use it directly:
https://time-machine-news-alpha.vercel.app/mcpIn Claude, open Settings, then Connectors.
Choose Add custom connector, name it Time Machine News, paste the link, and select Add.
In a chat, open the tools menu below the message box and turn on Time Machine News.
No account or API key is needed. To add it to Claude Code instead:
claude mcp add --transport http time-machine-news https://time-machine-news-alpha.vercel.app/mcpRelated MCP server: historical-investigator-mcp
Tools
Tool | What it does |
| Full-text search of page OCR, filtered by date range, state, newspaper (LCCN), and front pages only. |
| Front pages printed on a given day, with their opening text. |
| Full OCR text of one page, in chunks, with a citation and links to the scan and PDF. |
| Newspapers by name, state, or city, with their years in print and LCCN. |
Things to try: "What did Kansas newspapers report the week the Titanic sank?" or "Summarize the front pages from November 11, 1918."
Run locally
Requires Python 3.10 or newer.
git clone https://github.com/ronakrupani/time-machine-news.git
cd time-machine-news
python3 -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"Over stdio (Claude Desktop, Claude Code). The time-machine-news command runs the server over stdio. For Claude Desktop, add this to claude_desktop_config.json, using the absolute path to your clone:
{
"mcpServers": {
"time-machine-news": {
"command": "/absolute/path/to/time-machine-news/.venv/bin/time-machine-news"
}
}
}For Claude Code:
claude mcp add time-machine-news -- /absolute/path/to/time-machine-news/.venv/bin/time-machine-newsOver HTTP (the same app that runs on Vercel, including the landing page):
uvicorn time_machine_news.app:app --reloadThe landing page is at http://localhost:8000 and the MCP endpoint at http://localhost:8000/mcp.
Tests:
pytestThe tests use recorded responses and never call loc.gov.
Deploy your own
The project deploys to Vercel as-is: pyproject.toml sets the entrypoint ([tool.vercel]) and lists the dependencies.
vercel deploy --prodThe Library of Congress allows about 20 API requests per minute per client and blocks for an hour past that, so the server caches results and caps itself at 15 lookups a minute. Without Redis, each server instance keeps its own cache and counter. To share them across instances, add an Upstash Redis database (directly or through the Vercel Marketplace) and set these environment variables in Vercel (see .env.example):
UPSTASH_REDIS_REST_URL=
UPSTASH_REDIS_REST_TOKEN=The Marketplace integration's KV_REST_API_URL and KV_REST_API_TOKEN names work too.
Notes
Page text is machine OCR of old microfilm and contains errors. Check the scan before quoting.
Coverage ends in 1963; most pages are from 1836 to 1922.
A new search can take up to a minute because loc.gov is slow on broad queries. Repeat searches come from the cache.
Newspaper scans and text come from the Library of Congress. This project isn't affiliated with the Library.
License
Available Tools
4 toolsfind_newspapersFind newspapersARead-onlyIdempotent
Find newspapers in the archive by name, state, or city.
Returns each paper's title, years of publication, place, and LCCN (use it as
newspaper_lccn in search_newspapers to search just that paper).
Inputs: city, name, page, state, +1 more. Returns: object
front_pages_on_dateFront pages for a dateARead-onlyIdempotent
List newspaper front pages published on a specific day.
Each result includes the opening text of the page (usually the top headlines) and a
page_url for get_page_text. Good for "what was the news on <date>" questions.
Inputs: date (required), page, state, per_page. Returns: object
get_page_textRead a newspaper pageARead-onlyIdempotent
Read the full OCR text of one newspaper page, with a citation and links to the scan and PDF.
Long pages are returned in chunks: if `truncated` is true, call again with
`offset` set to `next_offset`.
Inputs: page_url (required), offset, max_chars. Returns: object
search_newspapersSearch newspaper pagesARead-onlyIdempotent
Search the full text of historical US newspaper pages (1700s-1963).
Returns matching pages with the newspaper, date, place, a text snippet around the
match, and a page_url to pass to get_page_text. Add a date range and a state or
newspaper whenever possible; broad searches are slow.
Inputs: query (required), page, match, state, +5 more. Returns: object
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 4 tools
Each tool targets a distinct action: full-text search, date-based browsing of front pages, reading a page's OCR, and locating newspapers by metadata. search_newspapers and front_pages_on_date both return pages, but one is text-driven and the other date-driven, so overlap is minor and the descriptions clarify intent.
Three of four tools follow a clean verb_noun pattern (search_newspapers, get_page_text, find_newspapers), and the parameters described (page_url, offset) reinforce readability. front_pages_on_date breaks the convention with a noun-first name, a minor deviation.
Four tools cover the essential workflow of a newspaper archive (discover papers, search text, browse a date, read a page) without redundancy. It is slightly lean but each tool clearly earns its place.
The surface covers discovery, search, browsing, and retrieval, forming a coherent end-to-end flow with page_url linking search results to get_page_text. Minor gaps exist (e.g., no way to list all issues of a paper by date range or filter results beyond search), but agents can work around these.
Maintenance
Related MCP Connectors
Give your AI assistant access to the world’s news. Search articles in any language, track coverage over time, and research companies and events.
Hosted browser for agents: open pages, search the web, extract data, read PDFs and take screenshots.
Google search, news, maps, webpages, cited answers, research and ChatGPT/Gemini answers for agents.
Notes, tasks, habits, journal and 6 more tools Claude can read and change, with undo. Free.
Related MCP Servers
- AlicenseNot gradedqualityCmaintenanceProvides access to the Library of Congress (loc.gov) data, enabling AI agents to search and retrieve information from the world's largest library.11 npmMIT
- AlicenseAqualityDmaintenanceEnables searching and retrieving historical records from the Library of Congress, including newspapers, photos, maps, manuscripts, audio, and film, via the Model Context Protocol.101MIT
- AlicenseNot gradedqualityBmaintenanceEnables full-text search of digitized historical U.S. newspapers from the Library of Congress Chronicling America collection, returning publication metadata and page-scan image URLs with filters for date range and state.177 npmMIT
- AlicenseAqualityCmaintenanceProvides AI assistants with programmatic access to the National Library of Sweden's digitized historical newspapers, enabling full-text search, filtering, timelines, and image retrieval across over 2.2 million pages.51MIT