概览
What is Immich Photo Manager?
Immich Photo Manager is an MCP (Model Context Protocol) server that gives any AI assistant direct access to a self-hosted Immich instance for intelligent photo management. Designed for users whose Immich library has grown beyond manual management, it enables natural‑language search, organization, deduplication, and album curation without scripts or manual sorting.
How to use Immich Photo Manager?
Install as a Claude plugin (claude plugin add immich-photo-manager) or configure any MCP‑compatible client with the package entry point via uvx. Set environment variables IMMICH_BASE_URL and IMMICH_API_KEY. The server defaults to stdio transport; set MCP_TRANSPORT=http for Streamable HTTP.
Key features of Immich Photo Manager
- AI‑powered natural language search via CLIP
- Geographic albums combining GPS, CLIP, and temporal matching
- Metadata repair: timestamps, GPS inference, timezone correction
- Duplicate detection using perceptual hashing across imports
- Bulk rotation (90°/180°/270°) — non‑destructive and revertible
- People and face management: list, merge, reassign, view thumbnails
- Library health report: inventory, metadata quality, storage analysis
- Tags: create, apply, and bulk manage across assets
- Interactive HTML galleries with three themes and batch actions
Use cases of Immich Photo Manager
- Ask “find my sunset photos from Italy” — natural language photo search
- Say “create albums for all my trips” — auto‑curated geographic albums
- Fix misdated photos (noon/midnight) or infer missing GPS coordinates
- Detect re‑encoded duplicates from Apple Photos, Google Photos, or other imports
- Rotate an entire album at once; check library health with one command
FAQ from Immich Photo Manager
How does Immich Photo Manager differ from manual scripts or the Immich UI?
It replaces manual API calls, JSON parsing, and per‑file EXIF editing with natural‑language commands. For example, “find my sunset photos from Italy” replaces writing search queries, and “repair timestamps” batch‑fixes metadata where the UI would require one‑by‑one editing.
What are the runtime requirements?
A running Immich instance (self‑hosted, v1.90+), an Immich API key, Python 3.10+ with pip. The server runs locally; your photos never leave your server.
Where does my photo data live, and what transport protocols are supported?
All operations occur against your self‑hosted Immich instance. The server uses MCP stdio transport by default; set MCP_TRANSPORT=http for Streamable HTTP. No external cloud dependency.
Which AI clients have been tested or are compatible?
Tested with Claude Desktop, Claude Code, and LM Studio (Gemma 4). Compatible with any MCP client (Cursor, Windsurf, VS Code, Cline, Zed, etc.).
How does the server handle safety and destructive actions?
It shows findings and asks before acting — e.g., when deleting, rotating, or merging assets. Galleries include a Cowork Actions Panel for batch operations where users paste the CLI command after reviewing the selection.
