flickr-mcp

An MCP server that gives AI assistants authenticated access to your Flickr photo library. Tag photos, write descriptions, set titles, manage group memberships, and run a full automated processing workflow — all driven by AI, applied to your photos.

Works with any MCP-compatible AI client: Claude, Cursor, Windsurf, and others.

What it does

Once connected, you can ask your AI assistant things like:

• "Process all my untagged photos" — scans your whole library, tags and describes everything in one pass
• "Tag all the photos in my Japan album"
• "Write a description for this photo" (with a Flickr URL or photo ID)
• "Find my car photos and add them to relevant Flickr groups"
• "Look at my recent uploads and add relevant tags to each one"

The AI looks at your photos visually (via built-in image fetching), generates metadata from the actual image content and EXIF data, and either previews changes before applying or runs fully automated depending on the tool.

Requirements

• Node.js 18 or later
• A Flickr Pro account — Flickr requires Pro to register API applications
• An MCP-compatible AI client

Setup

1. Register a Flickr API key

Note: Flickr requires a Pro account to register API applications (~$8/month or ~$72/year). If you're using Flickr seriously enough to want AI-assisted tagging, you likely already have one.

Go to flickr.com/services/apps/create/noncommercial and fill out the short form — any app name works, select "personal use." It takes about two minutes. Copy the API key and API secret from the next page.

2. Run the setup command

npx flickr-mcp setup

A browser window will open with a setup wizard. Paste your API key and secret, then follow the prompts to authorize access to your Flickr account. Your credentials are saved to ~/.config/flickr-mcp/credentials.json and never leave your machine.

3. Add flickr-mcp to your AI client

The easiest way is to paste the following prompt directly into your AI assistant. It contains everything needed — the config entry, the setup reminder, and a pointer to the full docs:

Add an MCP server to my config with these details:

• Name: flickr
• Command: npx
• Args: ["flickr-mcp"]

The full config entry is:

"flickr": {
  "command": "npx",
  "args": ["flickr-mcp"]
}

Add this to the mcpServers block in my MCP config file, then let me know where you put it and remind me to restart the client.

Full documentation and source: https://github.com/MK3Core/flickr-mcp

Your assistant knows where its own config file lives and will place the entry correctly regardless of which client you're using. After it confirms, restart the client.

{
  "mcpServers": {
    "flickr": {
      "command": "npx",
      "args": ["flickr-mcp"]
    }
  }
}

{
  "mcpServers": {
    "flickr": {
      "command": "npx",
      "args": ["flickr-mcp"]
    }
  }
}

Full automated workflow

The quickest way to process your entire library is two tool calls:

1. flickr_get_unprocessed_photos — scans every photo in your library and returns a list of anything missing tags, a description, or a real title. Raw camera filenames like DSC_0042 or Lightroom export suffixes like IMG_1234-Edit are flagged automatically.

2. flickr_process_photo — applies title, description, tags, and group assignments to a single photo in one atomic call. The AI fetches the image visually, reads EXIF for camera/lens/location data, generates up to ~20 focused tags following Flickr best practices, writes a concise description, and adds the photo to relevant groups — joining them automatically if needed.

Just ask: "Process all my untagged photos" and the AI will page through the unprocessed list and handle each one.

Available tools

These are the tools your AI assistant can call once the server is running. You don't interact with them directly — just describe what you want in plain language.

Reading your library

Editing metadata

Group management

Automated workflow

Individual write tools (flickr_add_tags, flickr_set_metadata, flickr_add_to_group) show a dry-run preview before applying. flickr_process_photo runs without preview by default — it's designed for batch automation.

Tagging approach

The AI follows Flickr-specific best practices when generating tags:

• Target ~20 tags per photo — enough to reach the right audiences without diluting reach
• Broad + specific subject pairing: "dog" and "golden retriever", not just one or the other
• Location chain: continent, country, region, city, venue where available
• Gear tags from EXIF: camera body and focal length (e.g. "sony a7 iv", "85mm")
• Never: noise tags ("photo", "image", "picture"), irrelevant popular tags, or tag spam — Flickr's spam policy can hide photos from search

Privacy and scope

This server can only access your own photos. It has no ability to read, search, or modify other users' content. The write access it requests is limited to your account.

Your API credentials are stored locally at ~/.config/flickr-mcp/credentials.json and are never sent anywhere except Flickr's own API.

Rate limiting

The server uses a token bucket that targets 3,000 requests/hour (Flickr's limit is 3,600). Burst of up to 10 requests is allowed. Image fetches count against the budget. Under normal use you will not hit this limit.

Running from source

git clone https://github.com/MK3Core/flickr-mcp
cd flickr-mcp
npm install
npm run build
node dist/index.js setup

Then point your MCP client at node /path/to/flickr-mcp/dist/index.js instead of npx flickr-mcp.

Support

flickr-mcp is free and open source. If it saves you time, a GitHub sponsorship is always appreciated — it helps justify spending more time on projects like this.

Contributing

Issues and pull requests are welcome. The codebase is intentionally small — around 700 lines of TypeScript across six files.

src/
  index.ts          Entry point
  setup.ts          OAuth setup flow
  auth.ts           OAuth 1.0a implementation + credential storage
  rate-limiter.ts   Token bucket (3,000 req/hour)
  flickr-client.ts  Flickr API wrapper
  tools.ts          All tool definitions and handlers