langchain-serpdive 0.1.0

LangChain integration for SERPdive, the AI Search API and Tavily alternative: answer-ready web content for agents, same speed, 20.2% fewer tokens, higher answer quality (60.7% of decided duels) on a public benchmark

langchain-serpdive

The official LangChain integration for SERPdive, the AI Search API: your agent asks a question, the tool returns answer-ready web content that is extracted, cleaned, and sized for an LLM. On a public, replayable 1,000-question benchmark, SERPdive runs at the same speed as Tavily, feeds your LLM 20.2% fewer tokens, and wins 60.7% of decided quality duels.

Install

pip install langchain-serpdive

Get a free API key (no card required) at serpdive.com/dashboard/keys and set it:

export SERPDIVE_API_KEY="sd_live_..."

Use as a tool

from langchain_serpdive import SerpdiveSearch

tool = SerpdiveSearch()
tool.invoke({"query": "latest developments in solid state batteries"})

The output is a lean dict your model can quote and cite directly:

{
    "query": "latest developments in solid state batteries",
    "results": [
        {
            "url": "https://...",
            "title": "...",
            "date": "2026-07-11",
            "content": "The fact-carrying text of the page, extracted and cleaned...",
        },
        ...
    ],
}

Use in an agent

from langchain.chat_models import init_chat_model
from langgraph.prebuilt import create_react_agent
from langchain_serpdive import SerpdiveSearch

agent = create_react_agent(
    init_chat_model("claude-sonnet-5"),
    tools=[SerpdiveSearch()],
)
agent.invoke({"messages": [("user", "What changed in the EU AI Act this month?")]})

Options

SerpdiveSearch(
    model="moby",     # "mako" (default): key sentences, fast. "moby": full page text, deep research
    answer=True,      # also return a direct answer synthesized from the sources
    max_results=5,    # hard cap on delivered results, 1-10
    api_key="sd_live_...",  # defaults to SERPDIVE_API_KEY
)

Localization is automatic: the language of the query picks where we search. There is no country parameter to configure. Failed searches are never billed.

Links

• API reference
• Public benchmark (replayable end to end: same questions, same judge, your machine)
• Python SDK (this package is a thin layer over it)
• MCP server for Claude, Cursor and other MCP clients

License

MIT