Developers

You can use Truncated's stories in your own apps, feeds and AI tools. There are three ways in: RSS feeds, a JSON API and an MCP server. Free and anonymous: no key or sign-in. Published stories only, 20 per page, in real time. Please read the terms of use before you build on it.

RSS

The main feed has the latest stories: https://truncated.ai/feed.xml

Every category and topic has its own feed: add /feed.xml to the address of its page, for example https://truncated.ai/category/models-research/feed.xml. Find a topic on the Topics page. Pages for companies, products and people have no feed of their own.

The category feeds:

Each feed has up to 50 recent stories. An item has:

title: The headline.
link: The original article at its publisher.
description: The two summary sentences, and the editors' note when there is one.
content:encoded: The same as HTML, as the site shows it: the summary, the editors' note with its links, and a link to the publisher.
category: The story's category.
guid: A stable ID for the story.
pubDate: The story's date.

The main feed can also carry one sponsored item, labelled as such.

API

Three read-only JSON endpoints, with snake_case keys. Lists return 20 stories a page: ask for the next one with page=2, and use page, pages and total in the answer to see where you are. A page past the end returns the last page. Category IDs: models-research, ai-agents, ai-development, infrastructure-hardware, media-vision-audio, robotics-autonomy, business-industry, policy-regulation, safety-security, benchmarks-evaluation, open-source-community, ai-society.

`GET /api/feed`

The latest stories.

category (optional): Category ID, as list_categories returns it.
page (optional): Page number, 1 to 100000, 20 stories per page (default 1). A page past the end returns the last page.

curl "https://truncated.ai/api/feed?category=models-research&page=1"

The answer:

stories: Up to 20 stories.
total: Stories matching the request.
page: The page returned.
pages: Number of pages.
filters: The filters applied, after unknown IDs were dropped.
scope: window: the recent stories.
notices: Why a filter was dropped, if one was.
archive_matches: Whether older stories in the archive match too.
content_version: A number that changes whenever published stories change.
site_total: Published stories site-wide.
revenue: The sponsored post shown with the feed (sponsor, labelled, or null) and the sponsor disclosures of editors' notes (disclosures, by story ID).

`GET /api/search`

Search all published stories, newest first.

q (required): Search text, 1 to 200 characters.
page (optional): Page number, 1 to 100000, 20 stories per page (default 1). A page past the end returns the last page.

curl "https://truncated.ai/api/search?q=agents&page=1"

The answer:

stories: Up to 20 matching stories, newest first.
total: Matching stories.
page: The page returned.
pages: Number of pages.

`GET /api/story/{id}`

One published story by ID.

id (required): Story ID: the id of a listed story.

id=$(curl -s "https://truncated.ai/api/feed" | jq -r '.stories[0].id')
curl "https://truncated.ai/api/story/$id"

The answer:

story: The story.
latest: The 5 newest other stories.
disclosure: The sponsor disclosure of its editors' note (label, href, sponsor_names), or null.
share_url: Its short share link.

Story fields

Every story in an answer has these fields:

id: Story ID.
headline: The headline.
summary: Two summary sentences written by Truncated.
canonical_url: The original article's identity URL at its publisher.
primary_url: The current link to the original article.
published_at: The publisher's date, when known, else null.
discovered_at: When Truncated found the story.
first_published_at: When Truncated first published the story.
rss_guid: The story's guid in the RSS feeds.
rss_date: The story's date in the RSS feeds.
publisher_id: Publisher ID.
publisher_name: Publisher name.
publisher_host: Publisher host name.
og_description: The publisher's own description of the article, or null.
image: The story image Truncated serves under /media/ (id, kind, owner, width, height, alt), or null.
favicon: The publisher's icon, the same shape as image, or null.
classification: Its filing (category_id, topic_ids, entity_ids, content_type_id, taxonomy_version, method), or null.
legacy_category: Only on stories from before the current categories: the old category name.
note: The editors' note (text, created_at, updated_at), or null.
coverage: Other outlets' reports of the same story (url, publisher_id, publisher_name, headline, position, same_publisher_allowed).

MCP

AI tools that speak the Model Context Protocol can connect to https://truncated.ai/mcp (Streamable HTTP, stateless, JSON responses; POST JSON-RPC). No key or sign-in.

Tools

`latest_stories`

The newest published Truncated stories from the rolling recent window, newest first, 20 per page. Optionally only one category (IDs from list_categories). Each story has a headline, two summary sentences, its publisher, the original article URL and a Truncated permalink.

category (optional): Category ID, as list_categories returns it.
page (optional): Page number, 1 to 100000, 20 stories per page (default 1). A page past the end returns the last page.

`search_stories`

Search all published Truncated stories by text, newest first, 20 per page.

q (required): Search text, 1 to 200 characters.
page (optional): Page number, 1 to 100000, 20 stories per page (default 1). A page past the end returns the last page.

`get_story`

One published Truncated story by its ID (the `id` of a listed story).

id (required): Story ID: the id of a listed story.

`list_categories`

The categories every Truncated story is filed under, with their IDs for latest_stories.

Tool results list stories in a short form, with page, pages, total and an attribution field that holds the terms of use. A story in a result has:

id: Story ID, for get_story.
headline: The headline.
summary: Two summary sentences written by Truncated.
publisher: Publisher name and host.
canonical_url: The original article.
permalink: The story's page on Truncated.
discovered_at: When Truncated found the story.
published_at: The publisher's date, when known, else null.
category: Category ID and label, or null.
coverage_count: How many other outlets reported the same story.

Claude Desktop

Add Truncated as a custom connector: in Claude, open Connectors (under Customize, or Settings in older versions), choose Add custom connector and enter:

Name: Truncated
Remote MCP server URL: https://truncated.ai/mcp
Authentication: No sign in

Claude Desktop's claude_desktop_config.json only starts servers on your computer, so a remote server like this one goes in Connectors.

Cursor

Add this to ~/.cursor/mcp.json for all projects, or to .cursor/mcp.json in one project:

{
  "mcpServers": {
    "truncated": {
      "url": "https://truncated.ai/mcp"
    }
  }
}

Other MCP clients that support Streamable HTTP can use the same URL.

Limits

Rate limits: MCP: 30 requests a minute per client address; requests from Anthropic's published Claude connector addresses get 300 a minute per address. REST: 60 requests a minute per client address. Over the limit the request fails with HTTP 429; retry after a minute. Reads are cached and refreshed when stories change; when busy, a read fails with HTTP 503 and Retry-After.

API: 60 requests a minute from each client address (IPv6 addresses count by /64). Over that, you get HTTP 429 with a short text message. Wait a minute, then try again.
MCP: 30 requests a minute per client address; requests from Anthropic's published Claude connector addresses get 300 a minute per address. Every request counts, including initialize and tools/list. Over that, you get HTTP 429 with Retry-After: 60 and a JSON-RPC error.
HTTP 503 with Retry-After means we are busy. Wait that many seconds (usually 10) and try again. Over MCP, a busy read comes back as a tool error that says when to retry, and HTTP 503 without Retry-After means the MCP server is unavailable; try again later.

Terms of use

Using the data means accepting these terms. It is the same text that /llms.txt and every MCP result carry:

Data supplied by the Truncated API (truncated.ai). Free to use, including commercially, with attribution: show 'Data: Truncated (truncated.ai)' with a link, and link each story to its original publisher. Headlines, article text and images belong to their publishers; Truncated supplies summaries, categories and links. Do not resell the feed itself as a data product. Provided as is, without warranty; access may be rate-limited or changed.

Help

Questions or problems: email support@truncated.ai. This is a free service, so we can't promise an answer or a fix, and there is no service level.

For AI tools, /llms.txt sums up Truncated, this API and the MCP server in Markdown.