langchain-w2a 0.1.0

LangChain integration for the Web2Agent Protocol — turn any W2A-enabled website into LangChain tools

langchain-w2a

LangChain integration for the Web2Agent Protocol.

Turns any W2A-enabled website's declared skills into LangChain tools — one tool per skill, with typed input schemas built from the site's agents.json.

pip install langchain-w2a

---

What is W2A?

W2A is an open standard where websites declare their capabilities at /.well-known/agents.json. Instead of crawling 40–50 pages to understand what a site does, your agent reads one file.

robots.txt  →  what not to crawl  (1994)
sitemap.xml →  where pages are    (2005)
agents.json →  what a site can do (2026)

---

Quick start

import asyncio
from langchain_w2a import W2AToolkit
from langchain_openai import ChatOpenAI
from langchain.agents import AgentExecutor, create_tool_calling_agent
from langchain_core.prompts import ChatPromptTemplate

async def main():
    # Each W2A skill becomes a typed LangChain tool
    toolkit = await W2AToolkit.from_url("w2a-protocol.org")
    tools = toolkit.get_tools()

    for tool in tools:
        print(f"{tool.name}: {tool.description}")
    # w2a_protocol_validate_manifest: Validate an agents.json...
    # w2a_protocol_check_site: Check whether a website...
    # w2a_discover: Discover what a W2A-enabled website can do.
    # w2a_call_skill: Call a specific skill on a W2A-enabled website.

    llm = ChatOpenAI(model="gpt-4")
    prompt = ChatPromptTemplate.from_messages([
        ("system", "You are a helpful assistant with access to W2A tools."),
        ("human", "{input}"),
        ("placeholder", "{agent_scratchpad}"),
    ])

    agent = create_tool_calling_agent(llm, tools, prompt)
    executor = AgentExecutor(agent=agent, tools=tools, verbose=True)

    result = executor.invoke({"input": "Is stripe.com W2A enabled?"})
    print(result["output"])

asyncio.run(main())

---

Generic discover + call tools

If you don't know the site in advance, use the generic tools:

from langchain_w2a import W2ADiscoverTool, W2ASkillTool

tools = [W2ADiscoverTool(), W2ASkillTool()]

# Agent can now discover any site and call its skills
# Example agent reasoning:
# 1. Call w2a_discover("stripe.com") → see available skills
# 2. Call w2a_call_skill("stripe.com", "search_products", {"q": "..."})

---

Toolkit — one tool per skill

For a known site, the toolkit pre-binds the URL and creates dedicated tools:

from langchain_w2a import W2AToolkit

# Only public skills (no auth required) by default
toolkit = await W2AToolkit.from_url("mysite.com")

# Include auth-required skills (you handle auth via headers)
toolkit = await W2AToolkit.from_url("mysite.com", include_auth_required=True)

# Only specific skills
toolkit = await W2AToolkit.from_url("mysite.com", skill_ids=["search", "checkout"])

---

Requirements

• Python 3.9+
• langchain-core >= 0.2
• w2a >= 0.1.0 (the W2A Python SDK)

---

License

Apache 2.0 — same as LangChain and A2A.

w2a-protocol.org · github.com/Nijjwol23/w2a