Ask Claude or ChatGPT about news today and it runs a web search, gets back a wall of links, and burns thousands of tokens parsing pages that all describe the same story. We built NewsMCP to fix that. It's a news MCP server that returns deduplicated news events — not article dumps — in responses small enough that your agent's context window barely notices them.

NewsMCP is live at newsmcp.com. You can try it in the browser without creating an account, add it to Claude Code or Cursor in a single step, and add an API key when your agent needs more history or more results per call.

What is NewsMCP?

NewsMCP is a deliberately light, read-only news product for AI agents: five tools, one story object, no full article text. Any agent (or any person) understands it in seconds.

It's built on the Model Context Protocol, the open standard that lets AI assistants connect to external tools and data. As of 2026, MCP is the default way agents reach beyond their training data: the ecosystem has grown from about 50 servers at launch in November 2024 to thousands of servers.

What's been missing from that ecosystem is a reliable way for agents to get news. Generic web search MCPs return links. NewsMCP returns events: when multiple outlets cover the same story, our clustering groups them into a single deduplicated record with citations, so your agent reasons over what happened instead of re-reading forty versions of the same headline.

NewsMCP is a product of NewsCatcher, built on the same news infrastructure that powers our enterprise News API.

Why not just wrap the News API in an MCP server?

Because MCP isn't an API, and treating it like one produces a bad tool. This is the design decision the whole product hangs on.

An API serves production pipelines. Large response bodies, full article text, dozens of parameters, and exhaustive results are features — your code parses what it needs. An MCP server serves a language model mid-conversation, where every token in the response competes with everything else in the context window. A 200-article JSON payload doesn't make an agent smarter. It makes it slower, more expensive, and more likely to lose the thread.

So instead of exposing our News API through MCP, we designed a separate product around one question: what does a model actually need to answer a news question? The answer: the event, who reported it, when, and where to verify. Nothing else.

What tools does NewsMCP include?

Five tools, all read-only:

  • news — search synthesized story events, or fetch one by id. Filters for subject, event type, content type, sector, minimum number of independent newsrooms, start date, sorting, and number of stories.
  • get_story — expand a single story: full summary, classification, actors, and source links.
  • check_coverage — judge how widely and how independently a claim has been reported.
  • check_limits — report what your caller may do right now: calls left this hour, lookback ceiling, largest allowed page, concurrency. Free to call — it's exempt from every metering stage, so checking never costs you a call.
  • check_health — confirm the API is reachable.

news and get_story both return stories. Each story is a cluster of articles about the same event, deduplicated, with citations to the underlying sources.

How does NewsMCP keep responses token-efficient?

Two things keep responses small.

First, the unit. One story replaces the twenty articles that covered it, so the token cost of understanding an event doesn't scale with how many outlets chased it.

Second, the verbosity parameter, which controls how many source links come back. Headline, summary, actors, and labels are always present — those are what the model reasons over. The links are what pad the response:

  • compact — no source links. For scanning many stories and deciding what matters.
  • standard (default) — up to three links per story. Enough to answer and cite.
  • full — every link, uncapped. For "show me every source that covered this," at the cost of a much longer response on a large limit.

Responses default to markdown, with text and json available when you want to parse rather than read.

How do I try NewsMCP?

Three ways, in increasing order of commitment:

  1. The playground — newsmcp.com has a live, keyless playground. Real queries, no signup, rate-limited.
  2. One-click install — buttons on the site. The keyless tier works out of the box: 20 calls an hour, a 3-day lookback, up to 5 stories per call, one request in flight at a time. Plenty for trying it inside a real session.
  3. A free API key — a key doubles the lookback to 7 days, raises the ceiling to 20 stories per call, and lifts the rate limit to 50 calls an hour. Higher plans go further; check_limits always reports your live ceilings. Commercial use is permitted on every tier.

‍

When should I use NewsMCP vs the News API?

Use NewsMCP when an AI is in the loop: an assistant answering questions, an agent doing ad hoc research, a coding session that needs current events as context. It's optimized for exactly that — small, structured, citation-backed responses a model can use immediately.

Use the News API for production workflows: monitoring pipelines, sentiment models, RAG systems that need full article text, NLP enrichment, and archives back to 2019 across 140k+ sources. NewsMCP is not a replacement for the News API, and it isn't trying to be. It's the front door for agents; the API is the infrastructure behind production systems. An API key widens NewsMCP's window to a week on the free tier and further on paid plans; anything deeper — full text, NLP enrichment, archives back to 2019 — is News API territory.

‍

FAQ

What is a news MCP server?

A news MCP server is a tool built on the Model Context Protocol that lets AI assistants like Claude retrieve news on demand. Instead of relying on training data or generic web search, the agent calls the server mid-conversation and gets back current, structured news data it can cite.

Is NewsMCP free to use?

Yes. The playground requires no account, and the keyless tier works in your own agent with no signup: 20 calls an hour, a 3-day lookback, up to 5 stories per call. A free API key extends the lookback to 7 days, raises the limit to 20 stories per call, and lifts the rate to 50 calls an hour. Commercial use is permitted on every tier.

Who is behind NewsMCP?

NewsMCP is a product of NewsCatcher, the company behind the News API and CatchAll web search API. It runs on the same news infrastructure that serves NewsCatcher's enterprise customers.

‍

Try it here: https://newsmcp.com/
Docs: https://newsmcp.com/docs
‍Questions: support@newsmcp.com