photo-search

Photo Search

Connection Required — ALWAYS CHECK FIRST

Before doing ANYTHING else in this skill, call ping on the Immich MCP server.

• If ping succeeds -> proceed with the skill normally.
• If ping fails or the MCP tools are not available -> STOP. Do not continue. Tell the user:

Immich is not connected. This plugin needs a running Immich MCP server to work.

Run /setup-immich-photo-manager to configure your Immich connection. You'll need:

1. Your Immich server URL (e.g., http://192.168.1.100:2283)
2. An Immich API key (how to create one)
3. The MCP server configured (see /setup-immich-photo-manager)

Do NOT skip this check. Do NOT try to run any other tool first. Always ping, always block if it fails.

---

CRITICAL RULE: NEVER CREATE TEMPORARY ALBUMS

This is the most important rule of this skill. NEVER create albums as part of a search workflow.

The user has a curated library of real albums. Creating temporary albums pollutes their library with junk. Instead:

• If photos belong to a real album -> use that real album directly
• If photos are NOT in any album -> show them directly using get_thumbnails_batch
• NEVER call create_album from this skill. Not for "temp" albums, not for "search results", not for any reason.

---

Search Workflow (Step by Step)

Step 1: Parse user intent

Identify what the user is looking for. Determine which search dimensions apply.

Step 2: Search for matching photos

Use search_metadata and/or search_smart to find matching assets.

IMPORTANT — Immich EXIF location quirks:

• Immich stores cities as municipalities, not tourist names. "Tikal" does not exist as a city — it's in the municipality of "Flores" (state: "Petén", country: "Guatemala").
• "Lanzarote" does not exist as a city — look for municipalities like "Arrecife", "Yaiza", "Teguise", "Haría", etc. (state: "Canary Islands", country: "Spain").
• When a place-name search returns 0 results, try broader terms: search by state or country instead of city, then filter. Or use search_smart with the place name as a CLIP query.

Search strategy priority:

1. search_metadata(city=...) — fastest, most precise if the city name matches
2. search_metadata(state=...) or search_metadata(country=...) — broader, catches municipalities
3. search_smart(query="...") — AI/CLIP semantic search, catches things without GPS

Step 3: Find REAL matching albums

Call list_albums() and fuzzy-match album names/descriptions against the user's query.

Examples:

• User asks "photos of Tikal" -> match album "Tikal & Petén" (and possibly "Guatemala")
• User asks "fotos de la Barceloneta" -> match album "Barcelona — Barceloneta / Playa"
• User asks "Valle del Jerte" -> match album "Valle del Jerte & Hervás"
• User asks "Lanzarote" -> match albums containing "Lanzarote" in their name

Matching rules:

• Case-insensitive substring match on album name
• Also check album descriptions
• The query terms can appear anywhere in the album name (e.g., "Tikal" matches "Tikal & Petén")
• Include albums for broader regions if relevant (e.g., for "Tikal" also include "Guatemala")

Step 4: Collect asset metadata for the gallery

Two paths depending on whether a real album was found:

Path A — Real album found (preferred)

Use get_album(album_id) to get the full asset list. Extract id, originalFileName, and fileCreatedAt from each asset. This is the best path because the user curated these albums intentionally.

Path B — No matching album (orphan photos)

Use the search results directly — they already contain id, originalFileName, and fileCreatedAt for each asset. Do NOT create an album. Just show the photos.

Fetch thumbnails as base64 using get_thumbnails_batch(asset_ids=[...], size="thumbnail", limit=50). The Cowork viewer runs in an about: sandbox that blocks ALL external network requests, so thumbnails MUST be embedded as base64 data: URIs.

Step 5: Build and present the HTML gallery

1. Read the template: assets/viewer-template.html from the plugin root
2. Replace all {{PLACEHOLDERS}} with actual data
3. Write the HTML to the outputs directory
4. Present via computer:// link

For Related Albums ({{ALBUMS_JSON}}):

• Include ONLY real albums found in Step 3
• If no real albums matched, use an empty array []
• NEVER fabricate album entries

---

Search Capabilities

Dimension	MCP Tool / Parameter	Example
Visual/semantic	search_smart(query=...)	"sunset at the beach", "birthday cake"
Location (text)	search_metadata(city=..., state=..., country=...)	city="Barcelona"
Date range	search_metadata(taken_after=..., taken_before=...)	2023-06-01 to 2023-06-30
Camera/device	search_metadata(make=..., model=...)	make="Apple", model="iPhone 14 Pro"
File type	search_metadata(asset_type=...)	"IMAGE" or "VIDEO"
Favorites	search_metadata(is_favorite=true)	true

Query Translation

User says	Search strategy
"photos from my Italy trip"	search_metadata(country="Italy") + list_albums() to find Italy albums
"sunset photos"	search_smart(query="sunset")
"photos from last Christmas"	search_metadata(taken_after="2025-12-20", taken_before="2025-12-31")
"my best photos"	search_metadata(is_favorite=true)
"photos taken with iPhone"	search_metadata(make="Apple")
"videos from Barcelona"	search_metadata(city="Barcelona", asset_type="VIDEO")
"show me Tikal"	search_metadata(state="Petén") + search_smart(query="Tikal") + match album "Tikal & Petén"

---

Gallery HTML Generation

Template

Use the canonical template at assets/viewer-template.html. Read the template file, replace {{PLACEHOLDERS}} with actual data, and write the result.

Placeholder Rules

• {{PAGE_SIZE}}, {{PHOTO_COUNT}}, {{ALBUM_TOTAL}}: Should be plain integers (e.g. 20). The template uses parseInt() with fallbacks, so non-numeric values degrade gracefully (PAGE_SIZE defaults to 6, others to 0).
• {{ALBUM_NAME}}: Can contain any characters including apostrophes (e.g. "L'Hospitalet"). Safe in HTML contexts. The JS alt-text reads from document.title instead of re-injecting this placeholder, so apostrophes won't break JS.
• {{SEARCH_QUERY}}, {{IMMICH_URL}}: Can be any string.
• {{PHOTO_ENTRIES}}: Must be valid JS object literals, comma-separated.
• {{ALBUMS_JSON}}: JSON album objects. The template wraps them in [...].flat(), so you can pass any of these formats:
  • Comma-separated objects: {"id":"abc","name":"X","total":50},{"id":"def","name":"Y","total":30}
  • A JSON array: [{"id":"abc","name":"X","total":50}]
  • Empty string (no albums): the template produces [].flat() → [] and hides the section

Thumbnail Delivery Strategies

There are three strategies for delivering thumbnails to the gallery viewer, with different trade-offs. The strategy used depends on the user's Immich setup and the viewing context.

Strategy 1: Base64 Embedded (Default — Always Works)

The Cowork viewer runs in an about: protocol sandbox that blocks ALL external network requests (fetch, <img src>, etc.). Therefore, the default and always-safe strategy is to embed thumbnails as base64 data: URIs directly in the HTML.

Each photo entry in {{PHOTO_ENTRIES}} includes the full thumbnail data:

{src:'data:image/jpeg;base64,/9j/4AAQ...',id:'<asset-id>',name:'<filename>',date:'<ISO-date>'}

• src: Base64 data URI of the thumbnail (from get_thumbnails_batch, size=thumbnail, ~250px, ~15-25KB each)
• id: The Immich asset ID (for linking to Immich web UI)
• name: Original filename (displayed as label)
• date: ISO date string from the asset metadata

Always use size="thumbnail" (250px) — never preview (1440px). Thumbnails average ~18KB each, so 50 photos ≈ 0.9MB HTML file.

How to get thumbnails: Call get_thumbnails_batch(asset_ids=[...], size="thumbnail", limit=50). If more than 50 photos, call in batches of 50.

Limitations: HTML file size grows linearly with photo count (~18KB per photo). Not ideal for albums with hundreds of photos. Maximum practical limit is ~50 thumbnails per gallery file.

CORS-Enabled Direct URLs (Optional — Requires User Setup)

If the user has configured CORS headers on their Immich reverse proxy, the gallery HTML can use JavaScript fetch() with the x-api-key header to load thumbnails on demand, converting responses to blob URLs. This enables full URL-based delivery for any photos, not just albums.

This requires the user to configure their reverse proxy (Nginx, Caddy, Traefik, etc.) to return CORS headers. See the Post-Install: CORS Configuration section in /setup-immich-photo-manager for instructions.

Advantages: Works for any photos (albums or search results), tiny HTML file, true on-demand pagination, no shared links needed.

Limitations: Requires CORS configuration on the server side. Not available out of the box.

Strategy Decision Flow

Always use Base64 Embedded (Strategy 1) as the default:
  ├─ ≤50 photos → Embed all thumbnails as base64
  └─ >50 photos → Embed first 50 in batches, warn user about file size and total count

Note: CORS-enabled direct URLs (Strategy 3) are an opt-in enhancement for users who open galleries in a regular browser. They do NOT work inside the Cowork sandbox. Never mention strategy numbers or internal labels in user-facing output — just generate the gallery silently using the correct approach.

Template lazy loading

The first page (PAGE_SIZE photos) loads immediately. Subsequent pages use IntersectionObserver to set src from dataset.src only when scrolled into view. Pagination is manual via "Load more" button (no infinite scroll). This works with both base64 and URL-based strategies.

Albums JSON Format

{{ALBUMS_JSON}} — a JSON array of REAL albums:

{"id":"abc123","name":"Tikal & Petén","total":169},{"id":"xyz789","name":"Guatemala","total":392}

Comma-separated JSON objects — NO outer array brackets (the template adds [...]). If no real albums match, use empty string.

Generation Workflow (Concrete Example)

User: "show me photos of Tikal"

1. ping() -> OK
2. search_metadata(state="Petén", country="Guatemala") -> found 200+ assets
3. list_albums() -> scan names -> found "Tikal & Petén" (id: d6dd63d0, 169 photos), "Guatemala" (id: 8dde4bb1, 392 photos)
4. get_album(album_id="d6dd63d0") -> get asset list with IDs, names, dates
5. get_thumbnails_batch(asset_ids=[...], size="thumbnail", limit=50) -> base64 JPEG data for first 50 photos
6. Read assets/viewer-template.html
7. Replace placeholders:
   - {{ALBUM_NAME}} -> "Tikal & Petén"
   - {{ALBUM_TOTAL}} -> 169
   - {{SEARCH_QUERY}} -> "Tikal"
   - {{IMMICH_URL}} -> "https://your-immich-server.com"
   - {{PAGE_SIZE}} -> 20
   - {{PHOTO_COUNT}} -> 50 (limited to 50 for file size)
   - {{PHOTO_ENTRIES}} -> {src:'data:image/jpeg;base64,...',id:"abc",name:"IMG_001",date:"2023-06-15"}, ...
   - {{ALBUMS_JSON}} -> {"id":"d6dd63d0","name":"Tikal & Petén","total":169},{"id":"8dde4bb1","name":"Guatemala","total":392}
8. Write tikal.html to outputs (~0.9MB with 50 base64 thumbnails, total album has 169 photos)
9. Present computer:// link

When photos are NOT in any album

User: "show me sunset photos"

1. ping() -> OK
2. search_smart(query="sunset") -> found 35 assets (each has id, originalFileName, fileCreatedAt)
3. list_albums() -> no album name matches "sunset"
4. Read template
5. Replace placeholders:
   - {{ALBUM_NAME}} -> "Sunset Photos"
   - {{ALBUM_TOTAL}} -> 35
   - {{PHOTO_COUNT}} -> 35
   - {{PHOTO_ENTRIES}} -> entries built from get_thumbnails_batch (base64 src + id + name + date)
   - {{ALBUMS_JSON}} ->     <-- empty string, no real albums match
6. Write sunset-photos.html to outputs (~0.7MB with base64 thumbnails)
7. Present computer:// link

---

Result Presentation

When showing search results:

• Count first: "Found 147 photos matching your search"
• Real albums: "These photos are in your album 'Tikal & Petén' (169 photos)"
• Date range: "Spanning from June 2019 to June 2023"
• Visual preview: Always generate the HTML gallery viewer
• Action prompt: Suggest next steps (see more photos, explore related albums, etc.)

Pagination

Immich API returns paginated results. For large result sets:

• Fetch first page to get total count
• Report total to user before fetching all pages
• For browsing, show first page thumbnails and offer to load more

Advanced Search Patterns

Finding screenshots

No GPS data + screen-resolution dimensions + no lens/focal length EXIF.

Finding duplicates

Same date range across import sources. Compare by exact hash, timestamp + dimensions, or CLIP similarity.