immich-mcp

immich-mcp

An MCP server for Immich self-hosted photo management that provides AI-accessible tools for browsing, searching, organizing, and managing photo libraries with duplicate detection and safe deletion workflows.

Category
Visit Server

README

immich-mcp

MCP server for Immich photo management. Provides AI-accessible tools for browsing, searching, organizing, and managing your self-hosted photo library — including duplicate detection and safe deletion workflows.

Features

  • 40+ MCP tools covering assets, albums, search, people, tags, shared links, activities, and duplicates
  • 3 MCP prompts for guided workflows: duplicate review, album organization, and untagged photo discovery
  • Tool annotations (title, readOnlyHint, destructiveHint, idempotentHint) for safe AI-driven automation
  • Dry-run support on all destructive operations
  • Duplicate analysis with format priority scoring and protection for favorited/album-linked assets
  • Automatic retries via httpx.AsyncHTTPTransport (configurable via IMMICH_MAX_RETRIES)

Requirements

  • Python 3.11+
  • Immich server with API access
  • uv (recommended) or pip

Installation

git clone https://github.com/whitehara/immich-mcp
cd immich-mcp
uv pip install -e .

Configuration

Set environment variables before running:

Variable Required Default Description
IMMICH_BASE_URL Yes Immich server URL (e.g. https://photos.example.com)
IMMICH_API_KEY Yes API key from Immich → Account Settings → API Keys
IMMICH_EXTERNAL_URL No same as IMMICH_BASE_URL Public URL used for web_url links (useful when base URL is internal)
IMMICH_TIMEOUT No 30.0 HTTP timeout in seconds. The /api/duplicates endpoint on large libraries (70k+ assets) can exceed 30 s — set to 120.0 or higher if immich_duplicates_list times out.
IMMICH_MAX_RETRIES No 3 Retry attempts on transient errors

Immich API Key Permissions

Immich 1.138.0+ supports granular API key permissions. Grant the following scopes when creating your key (Immich → Account Settings → API Keys → Create).

For full functionality (all tools enabled):

Scope Required by
server.about immich_ping
user.read immich_user_me
asset.read immich_assets_list/get/view, all search tools
asset.upload immich_assets_upload
asset.update immich_assets_update, immich_assets_bulk_update
asset.delete immich_assets_delete, immich_duplicates_delete
asset.statistics immich_assets_statistics
album.read immich_albums_list/get
album.create immich_albums_create
album.update immich_albums_update
album.delete immich_albums_delete
albumAsset.create immich_albums_add_assets
albumAsset.delete immich_albums_remove_assets
person.read immich_people_list/get
person.update immich_people_update
person.merge immich_people_merge
person.statistics immich_people_statistics
tag.read immich_tags_list/get
tag.create immich_tags_create
tag.update immich_tags_update
tag.delete immich_tags_delete
sharedLink.read immich_shared_links_list/get
sharedLink.create immich_shared_links_create
sharedLink.update immich_shared_links_update
sharedLink.delete immich_shared_links_remove
activity.read immich_activities_list
activity.statistics immich_activities_statistics
activity.create immich_activities_create
activity.delete immich_activities_delete
duplicate.read immich_duplicates_list
duplicate.delete immich_duplicates_dismiss

For read-only use (no create/update/delete tools), grant only: server.about, user.read, asset.read, asset.statistics, album.read, person.read, person.statistics, tag.read, sharedLink.read, activity.read, activity.statistics, duplicate.read

On Immich versions prior to 1.138.0, select All permissions.

Usage

stdio (Claude Desktop / Claude Code)

export IMMICH_BASE_URL=https://photos.example.com
export IMMICH_API_KEY=your-api-key
immich-mcp --transport stdio

Claude Desktop config (claude_desktop_config.json):

{
  "mcpServers": {
    "immich": {
      "command": "immich-mcp",
      "args": ["--transport", "stdio"],
      "env": {
        "IMMICH_BASE_URL": "https://photos.example.com",
        "IMMICH_API_KEY": "your-api-key"
      }
    }
  }
}

HTTP (Docker / remote)

docker run -e IMMICH_BASE_URL=https://photos.example.com \
           -e IMMICH_API_KEY=your-api-key \
           -p 8000:8000 \
           ghcr.io/whitehara/immich-mcp:latest

Or build locally with docker compose up.

Default transport in Docker is streamable-http on port 8000.

Tool Reference

Annotation columns: R = readOnlyHint, D = destructiveHint, I = idempotentHint

Each tool also carries a human-readable title annotation used by MCP clients for display.

Health

Tool Description R D I
immich_ping Verify connectivity, returns server version
immich_capabilities List supported server features

User

Tool Description R D I
immich_user_me Get the authenticated user's profile (name, email, quota, role)

Assets

Tool Description R D I
immich_assets_list List assets with filters (favorite, archived, trashed, type)
immich_assets_get Get full metadata for a single asset
immich_assets_view Get thumbnail, original, and web UI URLs for an asset
immich_assets_upload Upload an asset from a local file path or URL
immich_assets_update Update favorite, archived, description, rating
immich_assets_bulk_update Bulk update multiple assets (dry_run supported)
immich_assets_delete Delete assets, optionally permanently (dry_run supported)
immich_assets_statistics Get counts by asset type

Search

Tool Description R D I
immich_search_metadata Filter by date, type, location, camera, person, filename
immich_search_smart Semantic/CLIP search with natural language queries
immich_search_explore Discover popular places, people, and things

Albums

Tool Description R D I
immich_albums_list List all albums
immich_albums_get Get album details and assets
immich_albums_create Create a new album
immich_albums_update Update album name, description, or cover
immich_albums_delete Delete an album (assets are not deleted)
immich_albums_add_assets Add assets to an album
immich_albums_remove_assets Remove assets from an album

People

Tool Description R D I
immich_people_list List recognized people
immich_people_get Get person details
immich_people_update Update name or visibility
immich_people_merge Merge two face clusters
immich_people_statistics Get asset count for a person

Tags

Tool Description R D I
immich_tags_list List all tags
immich_tags_get Get a tag by ID
immich_tags_create Create a tag (use / for nested tags)
immich_tags_update Update tag name or color
immich_tags_delete Delete a tag

Shared Links

Tool Description R D I
immich_shared_links_list List all shared links
immich_shared_links_get Get shared link details
immich_shared_links_create Create a shareable URL
immich_shared_links_update Update expiry, password, permissions
immich_shared_links_remove Revoke a shared link

Activities

Tool Description R D I
immich_activities_list List comments and likes
immich_activities_create Add a comment or like
immich_activities_delete Delete an activity
immich_activities_statistics Get comment count

Duplicates

Tool Description R D I
immich_duplicates_list List duplicate groups with analysis (format score, resolution, protection status). Paginated (page/page_size, default 50 groups/page). Results are served from an in-memory cache; first call after server start returns cache_ready=false while data loads — large libraries may take 20–30 minutes on first load.
immich_duplicates_delete Delete assets from duplicate groups (dry_run supported)
immich_duplicates_dismiss Dismiss duplicate groups without deleting files (dry_run supported)

Pagination

immich_duplicates_list returns paginated results to avoid oversized responses on large libraries:

Parameter Default Description
page 1 Page number (1-based)
page_size 50 Groups per page (max 500)

The response includes total_groups, total_pages, page, and page_size fields. Iterate page=1..total_pages to process all groups.

Duplicate Analysis Fields

immich_duplicates_list with analyze=true (default) returns per-group analysis:

Field Description
keep_id Recommended asset to retain (highest quality)
safe_to_delete_ids Assets not favorited and not in any album
protected_ids Assets that are favorited or in albums — require explicit user approval
needs_review true when protected assets appear among delete candidates

Quality Priority for keep_id (highest to lowest)

Live Photo > Format score > Resolution > File size

Favorites and album membership are tiebreakers only — they do not override a quality difference. A favorited JPEG will not be recommended over an unfavorited RAW; instead the JPEG is listed in protected_ids for user review.

Format Score (highest to lowest)

RAW/DNG > HEIC/HEIF > PNG/TIFF > JPEG > WebP > GIF

Prompt Reference

Invoke prompts from your MCP client to start guided workflows.

immich/review_duplicates

Step-by-step workflow for reviewing and safely resolving duplicate photos:

  1. Fetch duplicate groups with analysis
  2. Present summary: safe-to-delete count vs. protected assets needing review
  3. For each group, choose an action:
    • Delete: keep the best asset, move the rest to trash (or permanently delete)
    • Dismiss: assets are intentionally kept as separate copies — remove the duplicate grouping without deleting files
  4. Dry-run confirmation before execution

immich/organize_album

Guided album creation workflow:

  • Organize by date, event, person, or location
  • Uses both metadata search and semantic (CLIP) search
  • Checks for existing albums to avoid duplicates
  • Confirms asset count before creating

immich/find_untagged

Discover unorganized assets in the library:

  • Identifies assets with no album, no description, or no recognized people
  • Suggests descriptions from EXIF data
  • Hands off to immich/organize_album for album creation

Safety Model

All destructive tools (delete, merge, remove) carry destructiveHint=true in their MCP annotations, signalling to clients that human confirmation is required before execution. Additional safeguards:

  • dry_run=true is the default on all delete operations — no changes occur without explicit opt-in
  • Duplicate analysis places favorited and album-linked assets in protected_ids; they are never auto-deleted
  • Prompts explicitly instruct the AI to present a summary and obtain user approval before any destructive action

Development

# Install with dev dependencies
uv pip install -e ".[dev]"

# Run tests
pytest

# Run a single test file
pytest tests/tools/test_assets.py -v

# Type check (optional)
pyright src/

Recommended Servers

playwright-mcp

playwright-mcp

A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.

Official
Featured
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.

Official
Featured
Local
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.

Official
Featured
Local
TypeScript
VeyraX MCP

VeyraX MCP

Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.

Official
Featured
Local
graphlit-mcp-server

graphlit-mcp-server

The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.

Official
Featured
TypeScript
Kagi MCP Server

Kagi MCP Server

An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.

Official
Featured
Python
E2B

E2B

Using MCP to run code via e2b.

Official
Featured
Neon Database

Neon Database

MCP server for interacting with Neon Management API and databases

Official
Featured
Exa Search

Exa Search

A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.

Official
Featured
Qdrant Server

Qdrant Server

This repository is an example of how to create a MCP server for Qdrant, a vector search engine.

Official
Featured