Blog / AI agents

How to give your AI agent live news with MCP.

October 5, 2026 · 6 min read

Language models know the world up to their training cutoff. Ask one what happened this morning and it either guesses or runs a generic web search. A news MCP server gives your agent a proper tool for that. Here's how to connect one to Claude, Cursor or any other MCP client in a few minutes, and how to get good results from it.

Why an agent needs a news tool

The Model Context Protocol (MCP) is the standard way to give an AI agent tools it can call. Most agents already have a web search tool, so why add another?

• Web search isn't news search. Search engines rank pages for the whole web. Ask about "EV battery news" and you get explainers from 2023 next to today's headlines. The agent has to open pages and work out which ones are current.
• Duplicates waste context. One wire story republished by twenty outlets uses twenty results. A news tool that groups coverage returns it as one story with "20 sources", which is also a useful signal of how big the story is.
• Relevance is checked before the agent sees it. The agent gets fewer, better results, uses fewer tokens, and gives better answers.

Plainradar's MCP server exposes one tool, search_news. It runs the same live search as plainradar.com: it reads the best-matching news sites and blogs in 45 languages, groups duplicate coverage, ranks the stories and returns only those an AI check confirmed as relevant. A call usually takes a few seconds.

The quick way: sign in

Clients that support sign-in (OAuth) need only the address. Add https://plainradar.com/mcp as a remote MCP server and the client opens a Plainradar page where you press Allow. If you don't have an account yet, enter your email there and open the link we send: that creates your free account, which includes agent access. There's no token to copy.

• Claude (web, desktop, mobile): open Settings → Connectors, add a custom connector and paste the address.
• Claude Code: run claude mcp add --transport http plainradar https://plainradar.com/mcp, then /mcp and choose to authenticate.

The app then shows under Account → Agent tokens, where you can disconnect it at any time. If your client asks for a header instead, or you're writing your own agent, use a token as described next.

Step 1: create a token

Sign up at plainradar.com (the free plan includes agent access), open Account → Agent tokens and press Create token. Copy it right away, because it's only shown once. Treat it like a password and keep it out of shared repositories. You can revoke it from the same page at any time.

The connection details are:

Endpoint: https://plainradar.com/mcp
Header:   Authorization: Bearer YOUR_TOKEN

The Connect an agent button at the top of the site shows the same details, with your token already filled in once you're logged in.

Step 2: add it to your client

Claude Code

One command, run in your terminal:

claude mcp add --transport http plainradar https://plainradar.com/mcp \
  --header "Authorization: Bearer YOUR_TOKEN"

Then start Claude Code and run /mcp to check that plainradar is connected.

Cursor

Add the server to ~/.cursor/mcp.json (or .cursor/mcp.json in a project):

{
  "mcpServers": {
    "plainradar": {
      "url": "https://plainradar.com/mcp",
      "headers": { "Authorization": "Bearer YOUR_TOKEN" }
    }
  }
}

Claude Desktop and clients that only run local servers

If your client can only start local (stdio) servers, use the mcp-remote bridge. It needs Node.js. In Claude Desktop, open Settings → Developer → Edit Config and add:

{
  "mcpServers": {
    "plainradar": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://plainradar.com/mcp",
               "--header", "Authorization: Bearer YOUR_TOKEN"]
    }
  }
}

Restart the app and the tool shows up in the tools menu. The same command works in any client that takes a command line:

npx -y mcp-remote https://plainradar.com/mcp --header "Authorization: Bearer YOUR_TOKEN"

Your own agent

Any MCP SDK that speaks streamable HTTP can connect: point it at the endpoint and send the header. The server is stateless and returns JSON, so it works behind proxies and in serverless functions.

What the tool returns

Here's a real response for electric vehicle batteries, trimmed to one story:

{
  "query": "electric vehicle batteries",
  "pages": [{
    "from": "2026-10-04T20:06:18Z",
    "to":   "2026-10-05T20:06:18Z",
    "stories": [{
      "title": "BMW Opens New €1 Billion Battery Plant in Germany to Supply New BMW i3",
      "link": "https://electriccarsreport.com/2026/10/bmw-opens-new-e1-billion-battery-plant-...",
      "published": "2026-10-05T07:45:12+00:00",
      "feed": "Electric Cars Report",
      "summary": "BMW has officially opened its new battery production plant in Irlbach-Straßkirchen, ...",
      "sources": 2,
      "relevant": "yes"
    }],
    "more": true,
    "cursor": "4AVB58NmZD6qxnbf:day"
  }]
}

That search returned 11 stories that passed the relevance check, in about six seconds. Each story has its headline, link, source, publication time, summary, image and how many sources covered it.

Parameters

Prompts that work well

The agent decides when to call the tool, so you can just ask normally. Some examples:

• "What's new in artificial intelligence this week? Group it by company and link the sources." The agent calls with days: 7.
• "Brief me on today's space exploration news. Skip opinion pieces." The agent adds a filter.
• "What are French outlets saying about artificial intelligence today?" The agent sets language: "French" and summarises in English.
• "Before you write the market update, check today's news on interest rates and cite anything that changes the picture."

Tips: ask for sources, so the agent keeps the links. For broad topics, let it follow the cursor once or twice. A higher sources count usually means a bigger story.

Credits and limits

Agent access is included in the free plan and uses the same credits as the website: each new query uses one credit. Cursor pages and filters within that query are included. Every response includes credits_left, so your agent can keep track of usage. Calls are rate-limited per account to keep the service fast for everyone.