How agents reach the open web: search, read and map public sources, and get back evidence that keeps its sources.
Beta. New accounts start with 1,000 free credits.
What it does
- Reach
- Search, scrape, map, crawl and browse public sources from one REST and MCP surface.
- Evidence
- Research requests come back as source-aware evidence from web, academic, code, package and dataset sources.
- Structure
- Pages come back as markdown, metadata, links and deterministic extraction outputs.
- Defaults
- Free providers come first, paid search is off by default, and browser fallback needs an explicit opt-in.
The source catalog
From the live catalog, October 12, 2026: 15 connectors and 40 providers in 8 lanes.
- Free web search
- searxng_local, ddg_html, marginalia, wikipedia, wikidata, wayback_cdx, commoncrawl
- Sitemap, RSS, and docs discovery
- robots, sitemap, html_links, rss, docs_paths
- Open scholarly archives
- arxiv, crossref, zenodo, osf, doaj, pmc, biorxiv, medrxiv
- Academic workflows
- arxiv, pubmed, semantic_scholar, openalex, open_archives
- Package registries
- pypi, npm, crates, maven, nuget, rubygems, packagist, pub
- Dataset registries
- huggingface, openml, dataverse, figshare
- Code intelligence
- github
- Domain and identity OSINT
- domain_intel, username_search, email_osint
For builders
- Documentationclawifi.dev/docs
The REST API, the MCP server and the source lanes.
- Python packagepypi.org/project/clawifi
pip install clawifi
- MCP server sourcegithub.com/atakanelik34/clawifi-mcp
The public MCP client and server.
- Source catalogapi.clawifi.dev/sources/catalog
Every lane and provider, as JSON.