pydantic-scavio

Real-time Google search for Pydantic AI agents, powered by Scavio — a unified search API for AI agents covering 31 platforms (Google, YouTube, Amazon, Walmart, eBay, Target, Home Depot, Reddit, X, TikTok, TikTok Shop, Instagram, LinkedIn, Threads, Kuaishou, Zillow, Redfin, Booking.com, Tripadvisor, Airbnb, Yelp, Indeed, Glassdoor, the app stores, G2, Capterra, ad transparency, SEC EDGAR and Companies House) plus an Extract endpoint for any URL.

Scope: 1 endpoint, by design

This package is a curated subset of the Scavio API, not a wrapper around all of it. It ships exactly 1 of Scavio's 195 endpoints:

That is the whole surface, and it stays that way on purpose: one small, well-described tool an agent picks correctly every time beats a hundred it has to disambiguate between.

Credits: 1 credit per search, flat. Nothing here reaches the endpoints that cost more (YouTube transcripts are 5, most Instagram calls are 2, a LinkedIn job is 30), nor the body-priced ones whose cost depends on the request rather than a flat number (Walmart, Threads, Kuaishou and extract).

Reaching the rest of the API

Scavio covers 195 live endpoints: 194 across 31 platforms, plus extract, which reads any URL. The platforms are Google, YouTube, Amazon, Walmart, eBay, Target, Home Depot, Reddit, X, TikTok, TikTok Shop, Instagram, LinkedIn, Threads, Kuaishou, Zillow, Redfin, Booking.com, Tripadvisor, Airbnb, Yelp, Indeed, Glassdoor, the Apple App Store, Google Play, SEC EDGAR, Companies House, G2, Capterra, Google Ads Transparency and the Meta Ad Library. The most recent batch added 93 endpoints across 22 platforms plus extract — 21 of those platforms are new to Scavio and Walmart was rebuilt. None of them are reachable through this package, which stays at Google web search. Point a Pydantic AI agent at the hosted MCP server for all of them:

from pydantic_ai import Agent
from pydantic_ai.mcp import MCPServerStreamableHTTP

scavio = MCPServerStreamableHTTP(
    'https://mcp.scavio.dev/mcp',
    headers={'x-api-key': 'sk_live_...'},
)

agent = Agent('openai:gpt-5.5', toolsets=[scavio])

That server exposes 191 tools; a curated subset loads by default and the rest is one env var away. See the MCP docs.

Alternatively, call the scavio SDK directly — it is already a dependency here — and wrap whichever endpoints you need as your own tools: client.youtube.transcript(...), client.ebay.search(...), client.sec.filings(...), client.meta_ads.search(...). Reading an arbitrary URL is a top-level method, not a namespace: client.extract(url, format="markdown"), never client.extract.extract(...). See the SDK reference and the full endpoint list.

The newer verticals each have their own page: eBay, Target and Home Depot, Zillow and Redfin, Booking.com, Tripadvisor and Airbnb, Yelp, Indeed and Glassdoor, the Apple App Store and Google Play, G2 and Capterra, Google Ads Transparency and the Meta Ad Library, SEC EDGAR and Companies House, Threads and Kuaishou, plus Extract.

Install

pip install pydantic-scavio

Get a free API key (50 one-time signup credits, no monthly refill, no credit card) at dashboard.scavio.dev.

Usage

from pydantic_ai import Agent
from pydantic_scavio import scavio_search_tool

agent = Agent(
    "openai:gpt-5.5",
    tools=[scavio_search_tool(api_key="sk_live_...")],
    system_prompt="Search the web when you need current information.",
)

result = agent.run_sync("What changed in the latest Pydantic AI release?")
print(result.output)

Fixing parameters for the agent

Parameters you pass to the factory are fixed for all searches and hidden from the LLM's tool schema; parameters you leave unset stay available for the LLM to set per call.

# Always search from the US in English; the LLM only controls the query.
tool = scavio_search_tool(api_key="sk_live_...", country_code="us", language="en")

Sharing a client

from scavio import AsyncScavioClient
from pydantic_scavio import scavio_search_tool

client = AsyncScavioClient(api_key="sk_live_...")
tool = scavio_search_tool(client=client)

Result shape

The tool returns list[ScavioSearchResult]:

License

MIT