Operations

Changelog

Versioned release notes for webfetch.

#v0.1.0 — Launch

Initial public release.

#Federated search across 24 providers

  • Public-domain / CC sources: wikimedia, openverse, flickr, internet-archive, smithsonian, nasa, met-museum, europeana, library-of-congress, wellcome-collection, rawpixel, burst, europeana-archival.
  • Stock photography: unsplash, pexels, pixabay.
  • Music identity: itunes, musicbrainz-caa, spotify, youtube-thumb.
  • Web search: brave, bing, serpapi.
  • Fallback: browser (Playwright, consent-gated).

Default set is 19 providers, auto-skipping any without auth. Five opt-in for ToS-grey or paid use cases.

#License-first ranking

Canonical tags: CC0, PUBLIC_DOMAIN, CC_BY, CC_BY_SA, EDITORIAL_LICENSED, PRESS_KIT_ALLOWLIST, UNKNOWN. Ranking by license, then confidence, then resolution. Default licensePolicy: "safe-only" rejects UNKNOWN entirely.

#Attribution machinery

  • attributionLine — single-string credit, rendered-ready.
  • XMP sidecars on webfetch download (ExifTool / Photoshop compatible).
  • JSON sidecars with full ImageCandidate provenance.
  • buildAttribution() exported from @webfetch/core for custom rendering.

#Four front doors

  • CLI — webfetch binary with search, artist, album, download, probe, license, providers, batch, watch, config.
  • HTTP server — seven POST endpoints, bearer auth, structured JSON responses.
  • MCP server — stdio transport, seven tools, one-command install for Claude Code.
  • Chrome extension — loaded as unpacked, lets you right-click any image for license.

#IDE integrations

Copy-paste MCP config for Claude Code, Cursor, Cline, Continue, Roo Code, and Codex.

#Browser fallback

Consent-gated headless Chromium, four-tier fallback ladder (Bright Data → Camoufox → Rebrowser → Vanilla), per-target adapters for Google Images, Pinterest, Twitter/X, and generic pages. Every browser-sourced candidate tagged UNKNOWN and flagged viaBrowserFallback: true.

#Performance + reliability

  • SHA-256-keyed content cache.
  • Per-provider rate limits with backoff.
  • Independent try/catch per provider — one failure doesn't cascade.
  • providerReports surface exposes per-provider timing and skip reasons.

#Tooling

  • GitHub Action for build-time asset fetching.
  • Homebrew tap: brew install ashlrai/tap/webfetch.
  • npm: npm install -g @webfetch/cli.
  • Docker: ghcr.io/ashlrai/webfetch and webfetch-server images.
  • Curl installer: curl -fsSL .../install.sh | bash.

#Documentation

Shipped with this launch: the full docs site (you're reading it now). Covers install, providers, license model, per-IDE MCP setup, HTTP API, CLI, browser layer, cookbook (10 recipes), self-hosting, FAQ.

#Known limitations

  • Bing adapter targets the v7 API, which Microsoft retired mid-2025. Works for legacy v7 endpoints only.
  • Self-hosted browser tier is Vanilla Playwright only; Bright Data / Camoufox / Rebrowser tiers are webfetch Cloud or BYOK.
  • No Helm chart yet — see Self-hosting for a hand-written Deployment.
  • find_similar only works with serpapi; other reverse-image providers planned.

Roadmap and upcoming releases live in GitHub issues.