Search API GitHub repositories in 2026

The SDK stays. The address changes

A search API repository on GitHub is a client. What you pay for is the address it calls. Tavily's own README shows its client taking another base address, api_base_url="https://your-gateway.com/tavily". Point that setting at SpecStream and every client.search() in your code is one search at $0.001, with the reply in Tavily's shape.

from tavily import TavilyClient
from exa_py import Exa

# Before: TavilyClient(api_key=TAVILY_KEY)
tavily = TavilyClient(api_key=SPECSTREAM_KEY, api_base_url="https://api.specstream.ai/tavily")
print(tavily.search("latest AI research")["results"][0]["url"])

# Before: Exa(api_key=EXA_KEY)
exa = Exa(api_key=SPECSTREAM_KEY, base_url="https://api.specstream.ai/exa")
print(exa.search("latest AI research", num_results=10).results[0].url)

The move for each API, with its parameters and reply fields, is on Migrate to SpecStream: Tavily, Exa, Brave and SerpApi.

List the search-api repositories with GitHub's own API

GitHub's REST API searches repositories at GET /search/repositories. This call lists the repositories that carry the search-api topic, most stars first, and needs no token:

curl "https://api.github.com/search/repositories?q=topic:search-api&sort=stars&order=desc&per_page=20"

GitHub returns up to 1,000 results for each search, and allows 10 requests a minute without a token and 30 with one. The list holds search databases, self-hosted gateways and MCP servers next to the clients of paid APIs, so read each README for the address it calls and the key it needs before you build on it.

The most accurate web search API

SpecStream scores 97.9 F-score on all 1,000 official SimpleQA Verified questions (978 correct), and 64.0 item F1 and 43.4 row F1 on all 200 official WideSearch tasks. The Extract API scores 0.908 F1 on WCXB. Each score and its method is on the benchmarks page.

Search, pages and rows on one key

The Search API is GET /v1/web-search. The Extract API returns the main text of up to 20 pages or documents a call at $0.80 per 1,000 pages, and a page with no text is not charged. Deep Search returns every match on the live web as structured rows at 5¢, 15¢ or $1 a row, with the most a run can cost quoted before it starts. Agents connect to https://api.specstream.ai/mcp, set up as shown in the MCP docs.

SpecStream

The live web data engine. Its Search API answers the calls of the Tavily, Exa, Brave, Serper, SerpApi and Firecrawl clients at their own paths for $1 per 1,000 searches, and scores 97.9 F-score on all 1,000 SimpleQA Verified questions. The Extract API reads pages at $0.80 per 1,000, and Deep Search returns structured rows at 5¢, 15¢ or $1 a row.

Firecrawl MCP server

firecrawl/firecrawl-mcp-server carries the search-api topic and runs with npx -y firecrawl-mcp and a Firecrawl key. Its README says a search costs 2 credits, and Firecrawl's Hobby plan adds 1,000 credits per $5: $10 per 1,000 searches. SpecStream answers Firecrawl's search and scrape calls at https://api.specstream.ai/firecrawl at $1 per 1,000 searches.

Serper

Serper sells Google results in packs: 50k credits at $1.00 per 1,000, with 2,500 free queries to start. Clients on GitHub post to https://google.serper.dev. SpecStream answers the same call at https://api.specstream.ai/serper with the key in X-API-KEY, at $1 per 1,000 searches with no pack to buy.

search2ai

fatwang2/search2ai is a self-hosted search gateway for AI agents. It needs Node.js 22.18 or later and at least one search provider you configure, and it tries the next provider when one is rate limited. It also serves endpoints for existing Exa and Tavily clients. SpecStream answers those same clients with nothing to host, at $1 per 1,000 searches.

Searcharvester

vakovalskii/searcharvester starts with one docker compose up and serves a Tavily-compatible /search through SearXNG, with 100+ engines and no API keys for search, on a server you run. SpecStream answers Tavily's /search at https://api.specstream.ai/tavily for $1 per 1,000 searches, with nothing to host.