How do I convert web pages to Markdown in a LangChain agent?

Steps

1. Install langchain-strale and a chat-model package.
2. Create the toolkit with your Strale API key.
3. Bind url-to-markdown to your model.
4. Ask it to fetch a URL and run the tool call the model returns.

The solution

# pip install langchain-strale langchain-openai
from langchain_strale import StraleToolkit
from langchain_openai import ChatOpenAI

toolkit = StraleToolkit(api_key="sk_live_YOUR_KEY")
tools = [t for t in toolkit.get_tools() if t.name == "url-to-markdown"]

llm = ChatOpenAI(model="gpt-4o").bind_tools(tools)
result = llm.invoke("Fetch https://example.com and convert it to Markdown")

# The model picks the tool and its arguments; your code runs the call.
for call in result.tool_calls:
    tool = next(t for t in tools if t.name == call["name"])
    print(tool.invoke(call["args"]))

How it works

The tool fetches one page you name, extracts the main content and converts headings, lists, links and code blocks to Markdown. The answer has the page title, the markdown and a word_count.

It identifies itself in its user agent, follows the site's robots.txt and stops at pages that block automated access, rather than working around the block. Pages behind a login or a paywall cannot be read, and pages built mostly in the browser may come back incomplete.

bind_tools only lets the model propose a call. To have the model read the result and answer in words, put the same tools in one of LangChain's agent loops.

Tools in this guide