Give your LangChain agent live web search and real-time news

A LangChain agent is only as current as the model behind it. Ask about something that happened this morning and the model either says it does not know or, worse, guesses with confidence.

The fix is to give the agent tools that reach today's internet. This guide uses langchain-nestdaddy, a LangChain integration for NestDaddy's own search index and real-time news collection, to add:

Class	What it gives the agent
`NestDaddySearch`	Web search: title, url, snippet and domain of each result
`NestDaddyNewsSearch`	Real-time news by key words, with an optional country and time window
`NestDaddyReadPage`	The clean main text of any page or article, without menus or ads
`NestDaddyNewsRetriever`	News as LangChain `Document`s for RAG, optionally with the full article text

Everything below was run against the live API with LangChain 1.5.

Install

pip install -U langchain langchain-nestdaddy export NESTDADDY_API_KEY="your-key"

Get a key from the NestDaddy MCP page (there is a free trial). The same key works for the REST API, the MCP server and this package.

Try the tools on their own

from langchain_nestdaddy import NestDaddyNewsSearch, NestDaddySearch NestDaddySearch(max_results=2).invoke({"query": "model context protocol"})

{ "query": "model context protocol", "count": 2, "results": [ {"title": "Roadmap - Model Context Protocol", "url": "https://modelcontextprotocol.io/development/roadmap", "snippet": "Our plans for evolving Model Context Protocol", "domain": "modelcontextprotocol.io", "type": "web"}, {"title": "Model Context Protocol", "url": "https://en.wikipedia.org/wiki/Model_Context_Protocol", "snippet": "Protocol for communicating between LLMs and applications: The Model Context Protocol (MCP) is an open standard...", "domain": "en.wikipedia.org", "type": "wiki"} ] }

News works the same way. Keep the query to a few key words, the way a headline would say it:

NestDaddyNewsSearch(max_results=3, days=2).invoke({"query": "ChatGPT Wallet"})

That returned, a few hours after the announcement:

2026-11-13 23:06 cryptobriefing.com OpenAI introduces ChatGPT wallet for Pro subscribers 2026-11-13 23:00 cnbc.com OpenAI launches ChatGPT Wallet, allowing its 'dots' AI agents to make purchases for users 2026-11-13 23:48 investing.com OpenAI launches ChatGPT Wallet, letting AI agents pay for purchases

Add "country": "US" (or GB, IN, SG, ID and others) to limit news to one country. Local news works in the local language, for example {"query": "banjir Jakarta", "country": "ID"}.

Give an agent live search and news

from langchain.agents import create_agent from langchain_nestdaddy import NestDaddyNewsSearch, NestDaddyReadPage, NestDaddySearch agent = create_agent( "anthropic:claude-sonnet-5-5", tools=[NestDaddySearch(max_results=5), NestDaddyNewsSearch(days=7), NestDaddyReadPage()], system_prompt=( "Use news search for anything recent, web search for everything else. " "Read a page in full before you explain its details, and cite the urls you used." ), ) result = agent.invoke({"messages": [{"role": "user", "content": "What did OpenAI announce this week?"}]}) print(result["messages"][-1].content)

Any chat model with tool calling works: swap the model string for OpenAI, Gemini, a local model through Ollama, and so on.

The system prompt matters more than it looks. Without it, many models never call a tool for "this week" questions, because they do not know their knowledge is out of date. One routing line ("news for recent, web for everything else") fixes most of those answers.

RAG over today's news

When you want documents rather than a tool-calling loop, use the retriever:

from langchain_nestdaddy import NestDaddyNewsRetriever retriever = NestDaddyNewsRetriever(k=3, days=2, full_text=True) docs = retriever.invoke("ChatGPT Wallet") for d in docs: print(d.metadata["source_domain"], d.metadata["full_text"], len(d.page_content))

cryptobriefing.com True 2611 cnbc.com True 4328 investing.com False 67

With full_text=True each article is read in full (one extra API call per article). The third article could not be read as clean text, so the retriever kept its summary instead of dropping it: you still get k documents, and metadata["full_text"] tells you which ones are complete.

Feed docs into any LangChain chain or prompt as context. Because every document carries its url and published_timestamp, the model can cite sources and say how fresh they are.

Async

Every tool and the retriever support ainvoke. The retriever reads full articles in parallel:

docs = await NestDaddyNewsRetriever(k=5, full_text=True).ainvoke("AI regulation")

Errors you might see

API errors raise NestDaddyAPIError with the HTTP status and a hint: a bad key points you to the dashboard, a plan limit to the plans page. The key is stored as a SecretStr, so it never shows up in logs or reprs.

Prefer MCP?

If you use Claude, Cursor, VS Code or n8n instead of LangChain, the same tools are available as one remote MCP server. Setup for each client is on the NestDaddy MCP page.