bugrahans's picture
Upload 4 files
4aeb13f verified
Raw History Blame Contribute Delete
9.61 kB
from __future__ import annotations
import json
import os
import urllib.parse
from typing import Any, Dict, List, Optional
import requests
from smolagents.tools import Tool
SCRAPE_DO_SIGNUP_URL = (
"https://dashboard.scrape.do/sign-up?utm_source=huggingface&utm_medium=google-search-scraper"
)
SEARCH_ENDPOINT = "https://api.scrape.do/plugin/google/search"
AI_OVERVIEW_ENDPOINT = "https://api.scrape.do/plugin/google/search/ai-overview"
# Credit costs from the Scrape.do docs.
CREDITS_PER_SEARCH = 10
CREDITS_PER_AI_OVERVIEW = 5
# (label, gl, hl, google_domain)
COUNTRIES: List[tuple] = [
("United States", "us", "en", "google.com"),
("United Kingdom", "gb", "en", "google.co.uk"),
("Germany", "de", "de", "google.de"),
("France", "fr", "fr", "google.fr"),
("Spain", "es", "es", "google.es"),
("Italy", "it", "it", "google.it"),
("Netherlands", "nl", "nl", "google.nl"),
("Turkey", "tr", "tr", "google.com.tr"),
("Canada", "ca", "en", "google.ca"),
("Australia", "au", "en", "google.com.au"),
("India", "in", "en", "google.co.in"),
("Japan", "jp", "ja", "google.co.jp"),
("Brazil", "br", "pt", "google.com.br"),
("Mexico", "mx", "es", "google.com.mx"),
]
COUNTRY_BY_GL = {gl: (label, gl, hl, domain) for label, gl, hl, domain in COUNTRIES}
TIME_PERIODS = ["last_hour", "last_day", "last_week", "last_month", "last_year"]
class ScrapeDoGoogleSearchTool(Tool):
name = "google_search_scraper"
description = """
Run a Google search and get the whole results page back as structured JSON:
organic results with positions, top and bottom ads, People Also Ask
questions, related searches, the AI Overview, knowledge panel, local map
pack, top stories, videos and forum discussions. Supports any country and
language, desktop or mobile layouts, time filters and pagination.
Powered by Scrape.do's Google Search API, which handles proxies, CAPTCHAs
and Google's changing HTML. Use this for rank tracking, SEO research,
ad monitoring, or grounding an answer in live search results.
"""
output_type = "string"
def __init__(self, api_token: Optional[str] = None) -> None:
self._api_token = api_token
self.inputs = {
"query": {
"type": "string",
"description": "The search query, exactly as you would type it into Google.",
},
"country": {
"type": "string",
"description": (
"Two-letter country code to search from, e.g. 'us', 'gb', 'de', 'tr'. "
"Sets the ranking country, interface language and Google domain. Defaults to 'us'."
),
"nullable": True,
},
"pages": {
"type": "integer",
"description": "How many result pages to fetch (1-5). Each page costs 10 credits. Defaults to 1.",
"nullable": True,
},
"device": {
"type": "string",
"description": "'desktop' or 'mobile' results layout. Defaults to 'desktop'.",
"nullable": True,
},
"time_period": {
"type": "string",
"description": (
"Only return results from a recent window: last_hour, last_day, "
"last_week, last_month or last_year."
),
"nullable": True,
},
"location": {
"type": "string",
"description": (
"City-level location in Google's format, e.g. 'Austin,Texas,United States' "
"or 'Istanbul,Istanbul,Turkey'. Must match the country code."
),
"nullable": True,
},
}
super().__init__()
# ------------------------------------------------------------------ token
def _resolve_token(self) -> Optional[str]:
if self._api_token:
return self._api_token.strip()
env_token = os.getenv("SCRAPE_DO_API_TOKEN")
return env_token.strip() if env_token else None
# ------------------------------------------------------------------ calls
def _get(self, url: str, params: Dict[str, Any]) -> Dict[str, Any]:
full_url = url + "?" + urllib.parse.urlencode(params, quote_via=urllib.parse.quote_plus)
try:
response = requests.get(full_url, timeout=120)
except requests.exceptions.RequestException as exc:
return {"_error": f"Request failed: {exc}"}
try:
data = response.json()
except ValueError:
return {"_error": f"Unexpected non-JSON response (HTTP {response.status_code})."}
if response.status_code == 401:
return {"_error": "Scrape.do rejected the token (HTTP 401).", "_auth": True}
if not response.ok:
message = data.get("message") or data.get("error") or f"HTTP {response.status_code}"
return {"_error": f"Scrape.do returned an error: {message}", "_status": response.status_code}
return data
def search(
self,
query: str,
country: str = "us",
pages: int = 1,
device: str = "desktop",
time_period: Optional[str] = None,
location: Optional[str] = None,
fetch_deferred_ai: bool = True,
) -> Dict[str, Any]:
token = self._resolve_token()
if not token:
return {
"status": "error",
"errorMessage": "Missing Scrape.do API token.",
"hint": f"Get a free token at {SCRAPE_DO_SIGNUP_URL}",
}
if not query or not query.strip():
return {"status": "error", "errorMessage": "Enter a search query."}
gl = (country or "us").strip().lower()
_, gl, hl, domain = COUNTRY_BY_GL.get(gl, ("", gl, "en", "google.com"))
pages = max(1, min(int(pages or 1), 5))
base: Dict[str, Any] = {
"token": token,
"q": query.strip(),
"gl": gl,
"hl": hl,
"google_domain": domain,
"device": device if device in ("desktop", "mobile") else "desktop",
"resolveGoto": "true",
}
if time_period in TIME_PERIODS:
base["time_period"] = time_period
if location and location.strip():
base["location"] = location.strip()
combined: Dict[str, Any] = {}
organic: List[Dict[str, Any]] = []
credits = 0
pages_fetched = 0
warnings: List[str] = []
for page_index in range(pages):
# The docs say to send `start` from the very first request of a
# multi-page sweep so all pages come from one coherent session.
params = dict(base, start=page_index * 10)
data = self._get(SEARCH_ENDPOINT, params)
if "_error" in data:
if page_index == 0:
result = {"status": "error", "errorMessage": data["_error"]}
if data.get("_auth"):
result["hint"] = f"Check your token or create one at {SCRAPE_DO_SIGNUP_URL}"
return result
warnings.append(f"Page {page_index + 1}: {data['_error']}")
break
credits += CREDITS_PER_SEARCH
pages_fetched += 1
for item in data.get("organic_results") or []:
item = dict(item)
item["page"] = page_index + 1
item["overall_position"] = len(organic) + 1
organic.append(item)
if page_index == 0:
combined = data
# The AI Overview can come back "deferred". Its session key
# expires after 90 seconds, so fetch it before any more pages.
ai = data.get("ai_overview")
if (
fetch_deferred_ai
and isinstance(ai, dict)
and ai.get("state") == "deferred"
and ai.get("session_key")
):
follow = self._get(
AI_OVERVIEW_ENDPOINT, {"token": token, "session_key": ai["session_key"]}
)
if "_error" in follow:
warnings.append(f"AI Overview follow-up: {follow['_error']}")
else:
credits += CREDITS_PER_AI_OVERVIEW
combined["ai_overview"] = follow
if not (data.get("pagination") or {}).get("next"):
break
combined["organic_results"] = organic
combined["status"] = "success"
combined["pages_fetched"] = pages_fetched
combined["credits_used"] = credits
if warnings:
combined["warnings"] = warnings
return combined
# ---------------------------------------------------------------- forward
def forward(
self,
query: str,
country: Optional[str] = None,
pages: Optional[int] = None,
device: Optional[str] = None,
time_period: Optional[str] = None,
location: Optional[str] = None,
) -> str:
result = self.search(
query=query,
country=country or "us",
pages=pages or 1,
device=device or "desktop",
time_period=time_period,
location=location,
)
return json.dumps(result, ensure_ascii=False, indent=2)