auto-browser LangChain Integration

LangChain, LangGraph and CrewAI adapters for Auto Browser. They call a running Auto Browser controller over its MCP tool endpoint.

pip install "auto-browser-langchain[langchain]"   # LangChain / LangGraph
pip install "auto-browser-langchain[crewai]"      # CrewAI

AutoBrowserTool is one LangChain tool that takes an MCP tool name and its arguments, so the model can use every tool the controller serves:

from auto_browser_langchain import AutoBrowserTool

tool = AutoBrowserTool(base_url="http://localhost:8000", bearer_token=None)
print(tool.invoke({"action": "browser.observe", "arguments": {"preset": "text"}}))

AutoBrowserNode is a LangGraph node that opens a session and records the current URL and screenshot in the graph state.

AutoBrowserNode needs only httpx. AutoBrowserTool needs langchain-core: building it without the [langchain] extra raises an ImportError that says so.

When a tool call fails, the controller still answers HTTP 200 with isError: true. AutoBrowserTool returns that text as ERROR: ... so the model can read it. AutoBrowserNode raises AutoBrowserToolError instead, with the tool name, the error text and the controller's structured payload (payload, code, retryable), so a graph can tell a missing session from a bad selector:

from auto_browser_langchain import AutoBrowserNode, AutoBrowserToolError

try:
    state = await AutoBrowserNode().run({"session_id": "s1"})
except AutoBrowserToolError as exc:
    if exc.code == "unknown_session":
        ...  # open a new session

Full walkthroughs: LangChain / LangGraph · CrewAI