Figma Context MCP
Give your coding agent access to your Figma data. Implement designs in any framework in one-shot.
作者提供的说明(英文原文)
Photo management for self-hosted Immich: search, albums, duplicates, faces, PDF reports. 94 tools.
这三项永远不会合并成一个分数:意见、测量和人气是不同的东西。
我们还没有为它写通俗的介绍。下面是我们从注册中心读取到的事实。
选择你使用的 AI 应用,然后复制设置。
IMMICH_BASE_URL, IMMICH_API_KEY标为机密的值是你账户的钥匙。千万不要分享给别人,也不要粘贴到聊天里。
你需要
你用的是哪个应用?
在上方选择你的应用。我们会记住它,下次按钮会显示“安装到 …”。
打开 Claude Desktop → Settings → Developer → Edit Config,把下面的内容粘贴到 claude_desktop_config.json
{
"mcpServers": {
"immich-photo-manager": {
"command": "uvx",
"args": [
"immich-photo-manager"
],
"env": {
"IMMICH_BASE_URL": "your-value",
"IMMICH_API_KEY": "your-value"
}
}
}
}在终端中运行这条命令
claude mcp add immich-photo-manager --env IMMICH_BASE_URL=your-value --env IMMICH_API_KEY=your-value -- uvx immich-photo-manager运行前请先填写空着的值。
粘贴到项目中的 .cursor/mcp.json,或 ~/.cursor/mcp.json
{
"mcpServers": {
"immich-photo-manager": {
"command": "uvx",
"args": [
"immich-photo-manager"
],
"env": {
"IMMICH_BASE_URL": "your-value",
"IMMICH_API_KEY": "your-value"
}
}
}
}浏览器会先询问是否打开应用,然后应用会显示设置并请你确认。密钥的值会留空,由你自己填写。
粘贴到项目中的 .vscode/mcp.json
{
"servers": {
"immich-photo-manager": {
"type": "stdio",
"command": "uvx",
"args": [
"immich-photo-manager"
],
"env": {
"IMMICH_BASE_URL": "your-value",
"IMMICH_API_KEY": "your-value"
}
}
}
}浏览器会先询问是否打开应用,然后应用会显示设置并请你确认。密钥的值会留空,由你自己填写。
这个服务器在你的电脑上运行,而 ChatGPT 只能连接互联网上的服务器。请选择其他应用。
你的选择通过 Cookie 保存在这台设备上,无需账户。
这些设置是根据作者在注册中心的条目生成的。如果 README 的写法不同,请以 README 为准。
它在你的电脑上运行;安装后,工具列表会显示在你的 AI 应用中。
我们检查的是任何人都能看到的信息;我们没有阅读代码。不适用的检查项不计入总数。 · 检查于 2026年11月8日
注册名称已通过 GitHub 账户 drolosoft 验证,该账户拥有此仓库
我们还没有从 GitHub 读取这个仓库
我们还没有从 GitHub 读取这个仓库
它在你的电脑上运行,没有互联网端点
在你的电脑上运行,改用本地设置
它在你的电脑上运行;我们只能读取互联网端点的工具列表
与注册中心里这个分类中所有活跃的服务器比较。只有当发布者拥有该仓库时,星标数才计入。
它在 1 项适用的基础检查中通过了 1 项。我们没有阅读代码。在添加密钥或令牌之前,请先看看这个服务器要求哪些权限,并使用只具备所需权限的密钥。
你需要填写:IMMICH_BASE_URL, IMMICH_API_KEY
它在你的电脑上运行。你的 AI 应用会在需要时启动它,数据不会经过作者的服务器(除非它自己调用在线服务)。
GitHub 仓库最后更新于–,最新版本(2.0.11)于2个月前进入注册中心。
支持 MCP 的应用,例如 Claude Desktop、Claude Code、Cursor 和 VS Code。你的电脑上还需要有 Python (uv), Docker。
同一分类中有:Figma Context MCP, Claude Real Video, Figma MCP Server
同一分类中其他作者的服务器,按星标数排序。
如何看卡片:↓ 每周下载量i★ GitHub 星标i三大支柱:评价 · 检查 · 使用情况i更新 = 最新版本发布到注册中心的时间i
Give your coding agent access to your Figma data. Implement designs in any framework in one-shot.
Let any LLM watch a video locally — and search everything it has ever watched.
The Figma MCP server brings Figma design context directly into your AI workflow.
按原文从作者的仓库复制。PopNx MCP 没有审查或背书其内容。HTML、图片和脚本已被移除。链接会离开本站。
immich-photo-manager
MCP server for intelligent photo management with [Immich](https://immich.app): your self-hosted library, understood.
An unofficial community plugin by Drolosoft. It is not affiliated with, sponsored by or endorsed by Immich or FUTO; "Immich" is their name and is used here only to say which server the plugin works with. It talks to your Immich server through its public API.
If your Immich library has grown past what you can manage by hand, immich-photo-manager gives any AI assistant direct access to your instance: search, organize, deduplicate, and curate albums through natural conversation. Works with Claude, Gemma, or any MCP-compatible client. Runs locally and talks only to your Immich; your originals stay on your server (see what leaves your network).
Tested, not assumed. Every push runs 364 unit tests on CI. Every release is also run live against real Immich 2.7.5 and 3.1.0 (Docker, all 94 tools over the MCP protocol's legacy era, state re-read after each write) before it is tagged; both protocol eras (the legacy handshake and stateless 2026-07-28) are pinned on every push by SDK-free wire tests (
tests/test_raw_wire_eras.py). The kit is intests/live/, reproducible by anyone. The demos indoc/demos/are transcripts of real sessions, Demo 11 is this exact flow prompt by prompt, and Demo 12 runs the video frames and PDF photobook on a real clip. Demos 13 to 18 cover everything added in 2.x (library discovery, OCR and people search, memories and stacks, partners and downloads, asset notes, the Docker image) from real sessions. Details: How it's tested.
Say "create albums for all my trips" and watch it work:
GPS coordinates, CLIP visual search, and temporal matching, combined in one request to create dozens of curated albums. No scripts, no manual sorting.
pip (download)git clone https://github.com/drolosoft/immich-photo-manager.git
cd immich-photo-manager pip3 install -r src/requirements.txt # the plugin runs on your system python3
claude plugin marketplace add ./ claude plugin install immich-photo-manager
The install prints that two options are not set yet: the server URL and the API key. Inside a session, /plugin asks for them in a dialog (the key goes to your system's secure storage); from the shell you can pass them with --config immich_base_url=... --config immich_api_key=..., or leave both empty and let the setup below save them.
Open Claude Code (restart it if it was already open) and connect it to your Immich. Guided:
/setup-immich-photo-manager
It asks for your server URL and API key, checks them against the server, saves them, and shows your library numbers:
Or skip the guide and say it in one line (same thing underneath):
Update my Immich credentials to http://immich.local:2283 with API key <your API key>
Either way the credentials are saved for every session from then on; repeat to change server or key. Confirm any time with:
What Immich version am I connected to?
That's the whole install. Claude Desktop, Cowork or another MCP client instead of Claude Code? That is the plain MCP server without the skills: see Getting Started, route B.
One line, no reinstall:
cd immich-photo-manager && git pull # the clone you installed from
claude plugin marketplace update drolosoft-marketplace claude plugin update immich-photo-manager@drolosoft-marketplace
Then restart Claude Code. drolosoft-marketplace is the name the marketplace gets when you add it from the clone (claude plugin marketplace list shows it). Your saved credentials carry over.
After pulling a new version, run pip3 install -r src/requirements.txt again: 1.7.1 added the video (av) and PDF (fpdf2) libraries to the plugin's dependencies. On the uvx route, uvx --refresh immich-photo-manager --help once, then restart the client.
Nothing to set up again. The options, the server command and the tool names are the same as in 2.0.12, and saved credentials carry over. What changes:
/plugin, then the plugin's options), which keeps the key in your system's secure storage. If you connected through /setup-immich-photo-manager, that keeps working and you do not have to move anything.The plugin process runs on your machine and only talks to your Immich. But everything the assistant *reads* through it goes to the model you use: filenames, dates, EXIF, album lists, and, when you ask it to look at pictures, thumbnails (250px by default, 1440px previews on request). Originals are never fetched. With Claude that means those thumbnails leave your network; with a local model over MCP (LM Studio, Ollama) nothing does. Nothing is sent unless you ask for it: listing albums or fixing dates moves text only, "tell me what's in these photos" moves images.
A PDF report follows the same rule: export_pdf writes the file to disk on the machine running the server, and it is not sent anywhere unless you pass return_base64=true. The file goes where output_path says (default your Desktop); existing files are never overwritten. Frames that only go into the PDF never leave your machine and cost no tokens; only the frames you ask the model to look at do. When the assets carry GPS, the Places page draws a map with tiles from tile.openstreetmap.org, the only third-party call the server itself makes; pass map=false to skip it and keep everything inside your network.
python3 on a file inside the plugin (src/plugin_entry.py). There are no hooks and no background jobs, and nothing is downloaded or installed when it starts. The Python libraries are the ones you install yourself from src/requirements.txt.x-api-key header. upload_asset sends a local file you name to that server.tile.openstreetmap.org, and only when a PDF report draws its Places map. The request carries the tile coordinates of the area, so it tells that server roughly where the photos were taken. map=false skips it.unpkg.com, each pinned to one version with an integrity hash so the browser refuses a changed file, and its map tiles from OpenStreetMap, which shows those two servers the areas you look at. Thumbnails are embedded in the page, or loaded from your own Immich when you set up CORS.export_pdf and download_archive write where you say and never overwrite. Galleries and maps are HTML files written by the assistant. The duplicate report can read an import folder on disk to compare photos, and it removes files there only when you approve the exact list./setup-immich-photo-manager or update_credentials), the key passes through the model like any other message and is saved in two files only your user can read: config.json in the plugin's cache folder and ~/.immich-photo-manager/config.json.create_shared_link makes a link that anyone with the URL can open. It is created only when you ask for sharing; list_shared_links shows the ones that exist and delete_shared_link removes one.delete_assets moves assets to Immich's trash, which restore_assets undoes. It deletes for good only with force=true, and empty_trash is permanent too. The cleanup and duplicate skills show what they found and wait for your answer before they delete anything. The plugin never changes Claude's permission settings.You never edit config files after setup. The connection is managed in conversation:
| You say | What happens |
|---|---|
| "What Immich version am I connected to?" | Reports the server version and the URL it's talking to |
| "Update my Immich credentials to `https://photos.example.com` with API key `...`" | Validates the key against that server, hot-swaps the live connection, persists it, no restart |
| "Show my Immich connection" | URL + masked API key |
One connection at a time: to work with a second Immich (a test instance, a friend's server), say the *update* sentence again; say it once more to go back. Wrong URL or key? It tells you, and keeps the previous connection.
Try the full walkthrough: [Demo 11, Album Walkthrough](doc/demos/11-album-walkthrough.md). Read an album item by item, find who repeats, create a sub-album, tag and describe every photo.
The same plugin runs in [Claude Code](https://docs.anthropic.com/en/docs/claude-code): search your library, curate albums, and generate galleries right from the terminal.
![Split scr
只显示了开头部分;其余内容请在仓库中阅读