When building autonomous AI agents with CrewAI or LangChain, giving your agent a web search or scraping tool often results in two failure modes: Anti-bot blocks: The target site blocks standard Python requests or playwright instances with C...
When building autonomous AI agents with CrewAI or LangChain, giving your agent a web search or scraping tool often results in two failure modes:
Anti-bot blocks: The target site blocks standard Python requests or playwright instances with Cloudflare, DataDome, or Akamai 403s.
Context explosion: When a page succeeds, the agent ingests 200KB of raw HTML, consuming context limits and driving up token costs.
Here is a self-contained custom Tool implementation using the MESSORA API that returns clean markdown directly to your agent.
CrewAI Custom Tool
import os
import requests
from typing import Type
from pydantic import BaseModel, Field
from crewai.tools import BaseTool
class MessoraScrapeInput(BaseModel):
url: str = Field(..., description="The full HTTP/HTTPS URL of the page to scrape.")
only_main_content: bool = Field(default=True, description="When True, removes navigation and footer boilerplate.")
class MessoraScrapeTool(BaseTool):
name: str = "web_content_scraper"
description: str = (
"Extracts clean markdown from any web page, handling JavaScript rendering "
"and anti-bot mechanisms automatically."
)
args_schema: Type[BaseModel] = MessoraScrapeInput
def _run(self, url: str, only_main_content: bool = True) -> str:
api_key = os.environ.get("MESSORA_API_KEY")
if not api_key:
return "Error: MESSORA_API_KEY environment variable is not set."
try:
resp = requests.post(
"https://api.messora.dev/v1/extract",
headers={"Authorization": f"Bearer {api_key}"},
json={"url": url, "only_main_content": only_main_content},
timeout=30,
)
resp.raise_for_status()
data = resp.json()
return data.get("markdown", "No content returned.")
except requests.exceptions.RequestException as err:
return f"Scraping failed for {url}: {err}"
Using it with a CrewAI Agent
from crewai import Agent, Task, Crew
researcher = Agent(
role="Senior Market Analyst",
goal="Extract and synthesize competitive data from target URLs",
backstory="You research companies and pricing pages accurately using live web data.",
tools=[MessoraScrapeTool()],
verbose=True,
)
research_task = Task(
description="Analyze the pricing page at https://example.com/pricing and summarize the tiers.",
expected_output="A bulleted summary of available plans and monthly prices.",
agent=researcher,
)
crew = Crew(agents=[researcher], tasks=[research_task])
result = crew.kickoff()
print(result)
Why this improves agent performance
Token efficiency: Markdown output averages 75% fewer characters than raw HTML.
Reliability: TLS fingerprinting and stealth routing are handled at the gateway layer, preventing 403 Forbidden errors in long-running agent loops.