LangChain エージェントに CapSolver を追加 — 1 回のツール呼び出しで CAPTCHA を解決します

# pip install "capsolver-agent[langchain]" langchain-openai langgraph
# export CAPSOLVER_API_KEY=CAP-XXXXXX
# export OPENAI_API_KEY=sk-XXXXXX

import asyncio
import os
from langchain_core.tools import tool
from langchain_openai import ChatOpenAI
from langgraph.prebuilt import create_react_agent
from capsolver_agent.schema import create_executor
from _env import load_example_env  # demo only: reads repo-root .env

load_example_env()

# 1. Wrap CapSolver as one LangChain tool.
#    createTask + result polling are handled inside capsolver-agent.
executor = create_executor()  # key from CAPSOLVER_API_KEY

@tool
async def solve_captcha(captcha_type: str, website_url: str, website_key: str) -> dict:
    """Solve a captcha (reCaptchaV2 / reCaptchaV3 / cloudflare) and return its token.

    captcha_type: captcha family, e.g. reCaptchaV2
    website_url:  full URL of the page that loads the captcha
    website_key:  site key found on that page
    """
    result = await executor.execute(
        "solve_captcha",
        {"captcha_type": captcha_type, "website_url": website_url, "website_key": website_key},
    )
    if not result.get("success"):
        return {"ok": False, "error": result.get("error", "unknown error")}
    return {"ok": True, "token": result["solution"]["token"]}

# 2. Any OpenAI-compatible chat model works; the LLM decides when to call the tool.
llm = ChatOpenAI(model=os.environ.get("OPENAI_MODEL", "gpt-4o"), temperature=0)

# 3. Give the tool to a ReAct agent.
agent = create_react_agent(llm, [solve_captcha])

# 4. Run.
PROMPT = (
    "Solve the reCaptchaV2 at https://www.google.com/recaptcha/api2/demo "
    "with site key 6Le-wvkSAAAAAPBMRTvw0Q4Muexq9bi0DJwx_mJ- "
    "and report the first 40 characters of the token."
)
result = asyncio.run(agent.ainvoke({"messages": [("user", PROMPT)]}))
print(result["messages"][-1].content)

# pip install "capsolver-agent[langchain]" langchain-openai langgraph
# export CAPSOLVER_API_KEY=CAP-XXXXXX
# export OPENAI_API_KEY=sk-XXXXXX

import asyncio
import os
from langchain_openai import ChatOpenAI
from langgraph.prebuilt import create_react_agent
from capsolver_agent.langchain_tools import get_langchain_tools
from _env import load_example_env  # demo only: reads repo-root .env

load_example_env()

# CapSolver's ready-made LangChain BaseTools, backed by capsolver-core:
# solve_captcha, detect_captchas, solve_on_page, get_balance, get_supported_captchas
tools = get_langchain_tools(api_key=os.environ["CAPSOLVER_API_KEY"])

llm = ChatOpenAI(model=os.environ.get("OPENAI_MODEL", "gpt-4o"), temperature=0)

# The solver tools sit beside the browser / data / workflow tools your agent already uses
agent = create_react_agent(llm, tools)

result = asyncio.run(  # CapSolver tools are async, so use ainvoke
    agent.ainvoke(
        {
            "messages": [
                {
                    "role": "user",
                    "content": "Solve the reCaptchaV2 at "
                               "https://www.google.com/recaptcha/api2/demo with site key "
                               "6Le-wvkSAAAAAPBMRTvw0Q4Muexq9bi0DJwx_mJ- and report the "
                               "first 40 characters of the token.",
                }
            ]
        }
    )
)

print(result["messages"][-1].content)