NestDaddy
Products Docs Blog Dashboard Sign in Get API key
n8n

Build an n8n AI agent that knows today's news (MCP or plain HTTP)

n8n's AI Agent only knows what its model learned in training. Two ways to fix it: connect an MCP server so the agent searches the web and news by itself, or call a news API from an HTTP Request node in a scheduled workflow. Step by step, no code.

NestDaddy team · · 4 min read

n8n's AI Agent node is great at reasoning over what you give it. Ask it about something that happened this morning, though, and it can only guess: the model behind it stopped learning months ago.

This guide shows two ways to give an n8n workflow live news and web search:

  1. Agent mode: connect an MCP server to the AI Agent node. The agent decides when to search the news, search the web or read an article in full. Best for chat-style workflows and research tasks.
  2. Pipeline mode: a scheduled workflow calls a news API with the HTTP Request node, then an LLM summarises the results and sends them to Slack, Telegram or email. Best for daily briefs and keyword alerts.

Both use the same API key. The examples use NestDaddy, an independent search engine with its own web index and news collection. A free 7-day key (5,000 calls, no card) is enough to build and test both.

Option 1: AI Agent + MCP Client Tool

What you build

Chat Trigger  ->  AI Agent  (model: any chat model you use in n8n)
                     |-- MCP Client Tool  ->  NestDaddy (news, web search, read page...)

Steps

  1. Add a Chat Trigger node (or any trigger that passes a question).
  2. Add an AI Agent node and connect your chat model (OpenAI, Anthropic, Gemini, Ollama or any model you already use in n8n).
  3. On the agent's Tool input, add an MCP Client Tool node: - Endpoint: https://mcp.nestdaddy.com/mcp - Server Transport: HTTP Streamable - Authentication: Bearer Auth, with a credential that holds your key (or Header Auth with Authorization: Bearer YOUR_KEY)
  4. Leave "Tools to include" on all tools, or pick the ones you need (for a news assistant: news_search, country_news, read_page).
  5. In the agent's system message, tell it when to use the tools:
You answer questions about current events. Always use news_search for anything
that may have happened recently, and read_page to read the one or two most
relevant articles in full before answering. Mention the date of each fact and
list the article URLs you used.

Open the chat and ask something like "What changed in EU AI rules this week?". In the execution log you will see the agent call news_search, then read_page, then answer with sources.

Why MCP instead of hard-coded HTTP calls

  • The agent chooses which tool to call and what to search for, based on the question.
  • One node gives the agent several tools: news by key words, country news, web search, page reading and research papers.
  • The same server works in Claude, Cursor and VS Code, so you test prompts once and reuse them everywhere (Claude setup, Cursor setup).

Option 2: scheduled news brief with HTTP Request

What you build

Schedule Trigger (every morning)
   -> HTTP Request   (news search for your topics)
   -> AI Agent / LLM (summarise into a short brief)
   -> Slack / Telegram / Send Email

Steps

  1. Schedule Trigger: every day at 07:00 in your time zone.
  2. HTTP Request node: - Method: GET - URL: https://nestdaddy.com/api/v1/news/search - Authentication: Generic Credential Type → Header Auth, with header name X-API-Key and your key as the value - Query parameters:
Name Value Notes
q OpenAI Your topic, as key words
days 1 Only the last day
limit 15 Enough for a brief
country GB Optional: only news from one country (two-letter code)

The response has a results list; each item has title, url, source_domain, published_timestamp and description. 3. AI Agent (or a plain LLM node) with a prompt like:

Here are today's news results as JSON: {{ JSON.stringify($json.results) }}
Write a brief for a busy reader: at most 6 bullet points, each one sentence,
merge duplicates of the same story, and end each bullet with the source link.
  1. Slack, Telegram or Send Email node: send the agent's output.

Several topics? Put a small list in a Code node before the HTTP Request (one item per topic); n8n runs the request once per item, and a Merge or Aggregate node collects them again before the summary.

Keyword alerts instead of a daily brief

Change the schedule to every 30 or 60 minutes, set days to 1, and add an IF node that only continues when results is not empty. To avoid sending the same story twice, keep the URLs you already sent (n8n's workflow static data in a Code node, or a small table in your database) and filter them out.

Which option should you use?

You want Use
A chat assistant that answers questions about current events Option 1 (MCP)
A research step inside a bigger agent workflow Option 1 (MCP)
A fixed daily or hourly brief Option 2 (HTTP Request)
Alerts when a company, product or person appears in the news Option 2 with an IF node

Many workflows use both: a scheduled brief for the team, and an agent for follow-up questions.

Costs and limits

Every MCP tool call or HTTP request counts as one call; connecting the MCP server and listing tools is free. A daily brief over three topics uses about 90 calls a month, and an agent that reads two articles per question uses three or four calls per question. Plans start at $9.99 a month for 10,000 calls after the free 7-day trial: get a key.

Prefer code over nodes? The same daily brief in Python, with dedupe and Telegram, Slack or email delivery: Build a daily news digest bot in Python.

More guides