webfetch MCP

License-first image search and download for any MCP-speaking agent.

Every result ships with a structured license tag (CC0, CC_BY, CC_BY_SA, platform-license tags, EDITORIAL_LICENSED, ...), a confidence score, and a ready-to-render attribution line. Covers 25 providers, with 19 in the default set — no API key needed for Wikimedia Commons, Openverse, iTunes, MusicBrainz CAA, NASA, The Met, Library of Congress, Internet Archive, Wellcome Collection, and Burst.

Why webfetch MCP?

Most image-search tools give you URLs. webfetch gives you URLs plus the legal metadata needed to classify them as open/commercial reusable, platform-license, editorial-only, or exploration-only. The MCP layer exposes seven composable tools — search, specialize by artist/album, download with a hash, resolve an arbitrary URL's license, reverse-image-search, and triage a source page — so an agent can go from prompt to attributed asset in a single conversation turn.

• 25 providers, 19 in the default set and many requiring no key at all
• License-first ranking: CC0 floats to the top; heuristic-only results stay below 0.5 confidence
• Attribution always included: one attributionLine string, ready for a tooltip or credits footer
• Free tier needs no API key for Wikimedia and Openverse; full provider coverage adds optional keys per provider

Install

One-line (recommended)

curl -fsSL https://raw.githubusercontent.com/ashlrai/webfetch/main/install/install.sh | bash

Installs bun if missing, builds the server, symlinks webfetch onto $PATH, and merges the MCP block into ~/.claude/settings.json.

Claude Desktop / Claude Code

Add to ~/.claude/settings.json:

{
  "mcpServers": {
    "webfetch": {
      "command": "bun",
      "args": ["run", "/path/to/webfetch/packages/mcp/src/index.ts"]
    }
  }
}

Cursor

Add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "webfetch": {
      "command": "bun",
      "args": ["run", "/path/to/webfetch/packages/mcp/src/index.ts"]
    }
  }
}

Cline

Paste into Cline's MCP settings JSON (VSCode sidebar → Cline → MCP Servers):

{
  "mcpServers": {
    "webfetch": {
      "command": "bun",
      "args": ["run", "/path/to/webfetch/packages/mcp/src/index.ts"],
      "disabled": false,
      "autoApprove": ["search_images", "search_artist_images", "search_album_cover", "probe_page"]
    }
  }
}

Replace /path/to/webfetch with your clone path (default after installer: ~/.webfetch/repo).

Tools

| Tool | Description | Key params | |------|-------------|------------| | search_images | Federated search across 25 providers. Returns the top candidates with license + attribution. Does not download. For large batches, use CLI handoff: webfetch batch --jsonl. | query, providers[], licensePolicy, minWidth, minHeight | | search_artist_images | Specialized artist image search with kind-aware provider routing: portrait, album, logo, performing. | artist, kind, providers[] | | search_album_cover | Canonical album artwork via MusicBrainz CAA + iTunes + Spotify. Results are EDITORIAL_LICENSED. | artist, album | | download_image | Download a URL to local disk cache. 20 MB cap, content-type guard, SHA-256 hash, host blocklist enforced. | url, maxBytes, cacheDir | | fetch_with_license | Resolve license for any URL via host heuristics + page metadata (<link rel=license>, dc.rights, OG tags). | url, probe (also download bytes) | | find_similar | Reverse-image-search a reference URL. Requires SERPAPI_KEY. Returns leads — verify before shipping. | url, providers[] | | probe_page | Enumerate every <img> on a webpage with inferred dimensions and per-image heuristic license. Respects robots.txt. | url, respectRobots |

License tags

Results carry one of: CC0 · PUBLIC_DOMAIN · CC_BY · CC_BY_SA · UNSPLASH_LICENSE · PEXELS_LICENSE · PIXABAY_LICENSE · EDITORIAL_LICENSED · PRESS_KIT_ALLOWLIST · UNKNOWN (rejected by default). Use open-only when platform/editorial licenses are not acceptable.

Migration note: Unsplash, Pexels, and Pixabay are no longer reported as CC0. Agents should treat UNSPLASH_LICENSE, PEXELS_LICENSE, and PIXABAY_LICENSE as platform-license results, not Creative Commons/public domain results.

Batch handoff

MCP tools are meant for interactive agent calls. For hundreds or thousands of queries, have the agent hand off to the CLI:

webfetch batch --file queries.txt --jsonl --continue-on-error --candidates 3

Input lines are query or query<TAB>provider-a,provider-b. Each output line contains index, query, status, candidateCount, candidates, top, downloads, and provider diagnostics.

Auth / env vars

| Provider | Env var | Free tier? | |----------|---------|------------| | Wikimedia Commons | — | Yes (no key) | | Openverse | — | Yes (no key) | | Unsplash | UNSPLASH_ACCESS_KEY | Demo key rate-limited | | Pexels | PEXELS_API_KEY | Free tier available | | Pixabay | PIXABAY_API_KEY | Free tier available | | Spotify | SPOTIFY_CLIENT_ID + SPOTIFY_CLIENT_SECRET | Dev app required | | Brave Search | BRAVE_API_KEY | Free tier available | | SerpApi | SERPAPI_KEY | Required for find_similar | | Bright Data managed browser | BRIGHTDATA_API_TOKEN | Optional server/cloud fallback |

MusicBrainz CAA and iTunes require no key. Missing keys cause the provider to skip gracefully — other providers still respond.

Demo prompts

Try these in Claude Code or Cursor after installing:

Find me a CC-licensed portrait of Miles Davis I can use on an album liner page.

Search for album artwork for "Kind of Blue" by Miles Davis and show me the attribution line.

Probe the page at https://commons.wikimedia.org/wiki/File:Miles_Davis.jpg and tell me if I can use it commercially.

Requirements

• Node ≥ 18 or Bun ≥ 1.0
• No API key required for basic use (Wikimedia + Openverse cover most editorial needs)

Links

• Landing site + docs: getwebfetch.com
• One-line install guide: docs/QUICKSTART.md
• Provider coverage matrix: docs/PROVIDERS.md
• License policy details: docs/LICENSE_POLICY.md
• Per-agent install guides: Claude Code · Cursor · Cline
• Issues: github.com/ashlrai/webfetch/issues