unsplash

name

unsplash

description

Find photos on Unsplash by keyword and return properly-sized, attribution-ready image URLs for any use case — blog hero images, inline article images, social media cards (OG, Twitter, LinkedIn, Instagram, Pinterest), thumbnails, illustrations, etc. Use this skill whenever the user wants to source, find, search for, or pick a photograph or stock image — even if they don't say "Unsplash" explicitly. Triggers on phrases like "find an image of X", "I need a photo of X", "stock photo", "hero image", "photo for my blog", "social media image", "OG image", and similar.

Unsplash photo search

Search Unsplash for photographs by keyword and return a URL sized correctly for the user's intended use (blog post hero, social media card, thumbnail, etc.) along with the attribution snippet Unsplash requires.

Workflow

1. Get the access key. Look at ~/.unsplash first. If it's missing, walk the user through creating one (see "Obtaining an access key" below). Never prompt for it if the file already has a value.
2. Confirm the use case if it's ambiguous. "Blog post" could mean hero or inline; "social" could mean any of half a dozen formats. A quick "is this for X or Y?" is fine. If the user already said (e.g., "for an OG image"), skip this.
3. Search via GET /search/photos with appropriate orientation for the use case. Default to per_page=10 so the user has choices.
4. Present candidates to the user. If the agent can render images in its UI, embed the thumb URLs (200px wide, cheap). Otherwise list each with its description / alt_description, photographer name, and the links.html URL on unsplash.com so they can click through.
5. Once the user picks one, build the final URL at the right size (see "Size presets") and produce the attribution snippet.
6. Trigger a download event by GET-ing the chosen photo's links.download_location — this is required by Unsplash's API guidelines any time a user actually uses a photo. It's a no-op for the user but it matters for the photographer's stats and your app's compliance.

Reading the access key

cat ~/.unsplash 2>/dev/null

The file should contain just the access key on a single line. If it's empty or missing, fall through to "Obtaining an access key".

Obtaining an access key

If ~/.unsplash doesn't exist, tell the user:

I need an Unsplash access key to search for photos. It's free and takes about 2 minutes:

1. Go to https://unsplash.com/oauth/applications and log in (or create an account).
2. Click New Application, accept the API terms, and fill in a name and description (anything is fine — "Personal use" works).
3. On the application page, copy the Access Key (the public one — not the Secret Key).
4. Paste it back to me and I'll save it to ~/.unsplash for next time.

Note: new keys start in "demo mode" with a 50-requests-per-hour rate limit, which is plenty for casual use. You can apply for production (5000/hour) later if needed.

When the user provides the key, save it:

echo "PASTED_KEY" > ~/.unsplash
chmod 600 ~/.unsplash

Searching

Use GET https://api.unsplash.com/search/photos with the access key as a Client-ID header:

curl -s "https://api.unsplash.com/search/photos?query=coffee+shop&per_page=10&orientation=landscape" \
  -H "Authorization: Client-ID $(cat ~/.unsplash)" \
  -H "Accept-Version: v1"

Useful query parameters

• query — keywords (required)
• per_page — 1–30 (default 10)
• orientation — landscape, portrait, squarish
• color — black_and_white, black, white, yellow, orange, red, purple, magenta, green, teal, blue
• order_by — relevant (default) or latest
• content_filter — low (default) or high

Picking orientation from use case

• Blog hero, OG image, Twitter card, LinkedIn post → landscape
• Pinterest, Instagram portrait/story → portrait
• Instagram feed square → squarish
• Inline blog body / unspecified → leave unset

Size presets

Every photo has urls.raw — an Imgix URL you append params to. Build the final URL as:

${photo.urls.raw}&w=WIDTH&h=HEIGHT&fit=crop&crop=entropy&auto=format&q=80

Keep the existing ixid parameter that's already in raw — it's how Unsplash tracks views back to the photographer. Just append with &.

Use case	Recommended URL params
Blog hero	&w=1600&auto=format&q=80&fit=max
Blog inline	&w=800&auto=format&q=80&fit=max
OG image / Twitter card / LinkedIn share	&w=1200&h=630&fit=crop&crop=entropy&auto=format&q=80
Twitter/X header	&w=1500&h=500&fit=crop&crop=entropy&auto=format&q=80
Instagram feed (square)	&w=1080&h=1080&fit=crop&crop=entropy&auto=format&q=85
Instagram portrait	&w=1080&h=1350&fit=crop&crop=entropy&auto=format&q=85
Instagram story	&w=1080&h=1920&fit=crop&crop=entropy&auto=format&q=85
Pinterest pin	&w=1000&h=1500&fit=crop&crop=entropy&auto=format&q=80
YouTube thumbnail	&w=1280&h=720&fit=crop&crop=entropy&auto=format&q=85
Thumbnail / avatar	&w=400&auto=format&q=75&fit=max

For retina, append &dpr=2 (or generate a srcset varying w).

fit=max preserves the aspect ratio (good for inline/hero where dimensions aren't strict). fit=crop&crop=entropy lets Unsplash auto-crop to fixed dimensions using the most visually interesting region (good for social cards where dimensions are strict).

auto=format is always worth including — it serves AVIF/WebP to browsers that support it, JPEG otherwise.

Attribution

Unsplash's API terms require attribution to the photographer and to Unsplash on each rendered photo. After picking a photo, generate this snippet for the user:

HTML:

Photo by
<a href="https://unsplash.com/@USERNAME?utm_source=YOUR_APP&utm_medium=referral"
  >PHOTOGRAPHER_NAME</a
>
on
<a href="https://unsplash.com/?utm_source=YOUR_APP&utm_medium=referral"
  >Unsplash</a
>

Markdown:

Photo by
[PHOTOGRAPHER_NAME](https://unsplash.com/@USERNAME?utm_source=YOUR_APP&utm_medium=referral)
on [Unsplash](https://unsplash.com/?utm_source=YOUR_APP&utm_medium=referral)

Replace YOUR_APP with the application name the user set up on Unsplash (slugified). If unknown, leave a placeholder and tell the user to fill it in.

Pull PHOTOGRAPHER_NAME from photo.user.name and USERNAME from photo.user.username.

Triggering a download

Once the user has picked a photo and is going to use it, hit the download endpoint. It returns immediately and just increments the photographer's download counter:

curl -s "$(echo "$DOWNLOAD_LOCATION_FROM_RESPONSE")?client_id=$(cat ~/.unsplash)" > /dev/null

Or, equivalently, with the header:

curl -s "$DOWNLOAD_LOCATION_FROM_RESPONSE" \
  -H "Authorization: Client-ID $(cat ~/.unsplash)" > /dev/null

This is in Unsplash's guidelines and is required to stay in compliance. Don't skip it.

What to return to the user

After they pick a photo, give them a compact block with everything they need:

Photo: [description or alt_description]
By: [photographer name] (@username)

Image URL (sized for [use case]):
https://images.unsplash.com/photo-XXX?ixid=...&w=1200&h=630&fit=crop&crop=entropy&auto=format&q=80

Attribution (Markdown):
Photo by [Name](...) on [Unsplash](...)

Unsplash page: https://unsplash.com/photos/XXX

If the user wants multiple sizes (e.g., a blog with both a hero and an OG image), generate all of them from the same urls.raw.

Edge cases

• Empty results. If total is 0, suggest broader or different keywords.
• 401 Unauthorized. The key is invalid. Tell the user and walk through obtaining a new one — don't silently fail.
• 403 Forbidden. Almost always means rate-limited in demo mode. Tell the user the limit resets hourly, or suggest applying for production access.
• Sensitive queries. Unsplash's content_filter=low default already blocks NSFW results. For content explicitly aimed at younger audiences, bump to content_filter=high.
• Faces / people in cropped results. When using fit=crop for fixed dimensions, swap crop=entropy for crop=faces,entropy if the photo features a person — keeps faces in frame.

Rate limits

Demo mode is 50 requests/hour. Production is 5000/hour. Each search counts as one request; image fetches (images.unsplash.com) don't count. Check the X-Ratelimit-Remaining response header if you suspect issues.