anysearch (Python)

One interface for every web search API. anysearch is an adapter layer — think LiteLLM, but for search providers — that lets you call Octen, Exa, Parallel, Brave, Tavily, SerpAPI, Perplexity, Linkup, Firecrawl, Keenable, OpenAI, and more through a single function with a common set of parameters and a common response shape. It also ships a dependency-free stdio MCP server and optional FastAPI proxy server that adapt to whatever API keys you have configured.

import anysearch

# Auto-selects a provider based on which API keys are set in your environment.
resp = anysearch.search("best vector databases 2026", max_results=5)
for r in resp:
    print(r.title, r.url)

# Pin a provider and ask for richer content — same parameters everywhere.
resp = anysearch.search(
    "how do transformers work",
    provider="exa",
    include_content=True,
    highlights=True,
    include_summary=True,
)

Install

Install the released package from PyPI, or the latest main from GitHub:

pip install anysearch-sdk

# or from main
pip install "git+https://github.com/dhruv-anand-aintech/anysearch.git@main#subdirectory=python"
# or
uv add anysearch-sdk

The base install works for all providers (every adapter is REST-based on httpx). Use extras to also install a provider's official SDK (for anysearch.native(...)):

pip install "anysearch-sdk[exa,parallel]"
pip install "anysearch-sdk[all]"
uv add "anysearch-sdk[exa,tavily,brave]"

REST-only providers (brave, serper, you, jina, kagi, searchapi, google-pse, searxng, keenable, tinyfish, deepseek, brightdata, dataforseo, openai, anthropic, xai) need no extra dependency — the extra name still resolves so the selection syntax is uniform.

Providers

Unified parameters

Unsupported parameters are handled by on_unsupported ("warn" default, or "ignore" / "error"). Unknown keyword arguments are forwarded to the provider as extra.

Common response shape

search() returns a SearchResponse:

resp.provider          # "exa"
resp.query             # "..."
resp.answer            # synthesized answer, if any
resp.results           # list[SearchResult]
resp.urls              # [r.url for r in results]
resp.to_dict()         # JSON-serializable

r = resp.results[0]
r.title, r.url, r.snippet, r.text, r.summary, r.highlights, r.score, r.published_date, r.author, r.source
r.raw                  # the provider's raw result

Fallbacks, async, and the native escape hatch

from anysearch import AnySearch

client = AnySearch(provider="exa", fallbacks=["tavily", "brave"])
resp = client.search("...", mode="deep")

# Async
resp = await client.asearch("...")

# Drop down to a provider's official SDK (requires the extra installed)
exa = anysearch.native("exa")        # -> exa_py.Exa instance

MCP server

anysearch-mcp            # stdio MCP server
anysearch-mcp --probe    # also verify each provider with a tiny live query at startup

Tools exposed: search, list_providers, check_providers, and migrate_codebase (scans a repo for direct search-API usage and suggests unified replacements). See the repo root README for client configuration examples.

FastAPI proxy

The proxy is a small search gateway inspired by LiteLLM Proxy: bearer-key auth, provider routing/fallbacks, OpenAI-style /v1/models and /v1/search, native /search, health checks, and simple in-process request limits.

pip install "anysearch-sdk[proxy] @ git+https://github.com/dhruv-anand-aintech/anysearch.git@main#subdirectory=python"

export ANYSEARCH_PROXY_KEYS="dev-key"
export EXA_API_KEY="..."
anysearch-proxy

curl -H "Authorization: Bearer dev-key" http://localhost:4000/v1/models
curl -X POST http://localhost:4000/v1/search \
  -H "Authorization: Bearer dev-key" \
  -H "Content-Type: application/json" \
  -d '{"model":"exa","query":"best vector databases","max_results":3}'

Useful env vars: ANYSEARCH_PROXY_KEYS, ANYSEARCH_PROXY_ADMIN_KEYS, ANYSEARCH_PROXY_REQUIRE_AUTH=0, ANYSEARCH_PROXY_REQUEST_LIMIT, ANYSEARCH_PROXY_HOST, and ANYSEARCH_PROXY_PORT.