Skip to content
Image Asset Server

AI agents

Let an AI agent find and license images for your project

Updated · 2 min read

Short answer

Add the MCP server (claude mcp add --transport http images https://image.shep.bot/mcp) and ask for what you need: 'a hero image of a mountain lake and three feature icons for the landing page'. The agent calls search_images or build_image_set, which only return commercial-safe images by default, downloads the files and writes the credit lines into ATTRIBUTION.md.

What the agent sees

Every result carries its pixel size, orientation, kind (photo, painting, vector, icon…), a thumbnail URL, the licence with its verdict, an editorial flag and a ready-to-paste credit line. The server’s instructions tell the agent never to present a non-commercial, unverified or editorial image as cleared for advertising.

A whole page in one call

build_image_set takes a list of slots:

hero: misty forest at sunrise, landscape (1920px)
3 icons: upload, share, settings
background: paper texture
portrait of a potter at the wheel

It returns one pick per slot with alternatives, the file to use (the smallest at least as wide as the slot), suggested alt text, and exports: images.json, credits.html with <figure> captions, download.sh and ATTRIBUTION.md. Try it in the browser on the image set builder.

Without MCP

The same features are a REST API: GET /v1/search?q=…, GET /v1/images/{id}/download?width=1920 (a redirect to the file) and POST /v1/sets. See the API guide.

Step by step

  1. Connect the server

    Run claude mcp add --transport http images https://image.shep.bot/mcp, or add the same URL in Cursor, VS Code or any MCP client. For local file writes, run npx -y image-asset-server mcp instead.

  2. Describe the images

    Say what each image shows, its shape and its use: 'a wide hero image of a misty forest', 'a vertical poster of a Hokusai wave', 'SVG icons for upload, share and settings'.

  3. Let it pick and check

    The agent searches every source, keeps commercial-safe licences, and for a whole page calls build_image_set, which picks one image per slot with alternatives.

  4. Review the credits

    Check ATTRIBUTION.md or credits.html, and any image flagged editorial, before you publish.

Questions

Will the agent use images I can't use commercially?
Not by default: search_images and build_image_set return only 'yes' and 'yes-with-credit' verdicts unless the agent sets commercial_only=false because you said the project is non-commercial.
Does the hosted server download files for the agent?
No. It returns direct links to the files on the sources; the agent downloads them with curl. The local MCP server (npx) can save files itself with download_image.
Which tools does the server offer?
search_images, get_image, build_image_set and list_sources, plus download_image when it runs locally.

Try it

Live searches across every source, with a licence verdict on each result.

Sources

Practical information, not legal advice. AI agent? This guide is also available as Markdown.

Free images to start with

All collections →
  • Free SVG icons from open-source icon sets

    Free SVG icons from open-source sets (Material Symbols, Tabler, Phosphor, Lucide and more) via Iconify, with each set's licence and transparent backgrounds.

    14 images from 1 source

  • Free 4K nature photos: mountains, lakes, coasts

    Free high-resolution nature photos, at least 4K on the short side: mountains, lakes, coasts and forests, with the licence and credit line on every image.

    23 images from 4 sources · 22 at 1080 px or larger