Real-time news for AI agents

Your agent knows what happened today.

Unzoi gives your AI the world’s news as it breaks: each story once per language, with who reported it, when, and how widely, including reporting that never appeared in English. Your model writes the answer; Unzoi does the collecting and grouping. Add it to Claude, ChatGPT, Cursor or your own code with one line.

Free to start, no card. Every feature on every plan.

Today: a hosted API and MCP server. Running it in your own cloud or network is planned and not yet started. What is live

Story 20260910-0731-4f2a

Export licence rules tightened for lithography tools

34 outlets
engdeujpn
first seen 07:31last seen 19:40
41 articles → 1 story credits_charged: 1
An illustration of one story as the API returns it, not a measurement.

Why not just web search?

Web search gives your agent pages. Unzoi tells it what happened.

Keep your search tool for documentation and the rest of the web. For news it hands your agent pages, and on a breaking story many of them are the same report under different mastheads. Nothing in the results groups those copies together or says who had it first.

Unzoi returns the event instead: one record per story, with every outlet in our index that ran it, which one had it first, and when. Your agent spends its context on different stories rather than repeats, and can tell your user how far a story spread. How far is not the same as how many confirmed it: one agency report can run in hundreds of outlets.

The panel beside this text is not a picture. It is the same question asked both ways, against today’s news.

Try your own topic

What a search returns

  • 20 results
  • 14 are the same report
  • ? how many outlets ran it
  • ? which languages
  • ? who had it first

What Unzoi returns

  • 1 event
  • 34 outlets carried it
  • 3 languages
  • first reported 07:31
  • a link to every article

An example of the shape of one answer, not live data.

Watch it work

A real AI agent, answering from today’s news

Every few hours we give Claude one question and access to Unzoi, and publish exactly what happened: what it looked up, what came back, and the answer it wrote. Nothing is scripted and nothing is edited, beyond trimming long results to fit the page.

Your agent gets the same tools. It decides what to look up, Unzoi answers in a form it can reason over, and the answer it gives your user comes with its sources.

See earlier runs

What changed on chip export controls this week?

Tool call list_stories q: "chip export controls", mode: "hybrid", from: "2026-09-08"
Result 12 stories · top: 34 outlets, 3 languages · credits_charged: 3

Twelve distinct stories. The largest, on export licence rules for lithography tools, ran in 34 outlets across English, German and Japanese, first seen at 07:31.

An illustration of an agent calling the server, not a recorded session.

What it changes

Built for products that need to know what is happening

Start

Add it in one line

One command adds it to Claude Code; the JSON block does the same for any AI app that supports MCP, the standard way to give an assistant new tools. Building your own product? It is one web request, and the same key works for both.

  • Free to start: 1,000 credits a month, about 1,000 searches by exact words or 333 that also match by meaning
  • No feature held back for paid plans
  • The cost of every call is shown before and after it runs

Ready-made examples for common jobs

Claude Code

claude mcp add --transport http unzoi https://api.unzoi.com/mcp \
  --header "x-api-key: $UNZOI_API_KEY"

Claude, ChatGPT, Cursor

{
  "mcpServers": {
    "unzoi": {
      "url": "https://api.unzoi.com/mcp",
      "headers": { "x-api-key": "your-key" }
    }
  }
}

Your code

curl -s -G "https://api.unzoi.com/stories" \
  -H "x-api-key: $UNZOI_API_KEY" \
  --data-urlencode "q=port strike"

For teams

Monitoring, research, risk or a news feed? Start from your job

Teams use the same index to follow companies, markets and places: one alert per story rather than per article (a story reported in two languages still arrives twice for now), with which outlet had it first in our index. Your data engineer connects it and sets up the watches; your team keeps its own review workflow. Building a news feed into a product? The same API returns grouped stories with headlines, snippets, links and timestamps.

What it will not do: Unzoi tells you what was reported and by whom, not whether it is true. It returns headlines, snippets and links rather than full articles, which does not by itself settle what you may display or store, so check the terms for your use. Where it is the wrong tool What we have and have not measured

Why we are building this

An AI that answers questions about the world needs to know what happened today.

Models are frozen on the day their training ends, and the assistants, copilots and agents built on them are being asked about this morning. Web search was built for people reading pages, not for software that needs to know what happened. Unzoi is the layer in between: the world’s reporting, turned into events an AI can use.

Read why, and where it is going