markie-mcp

MCP (Model Context Protocol) server for Markie — a personal asset library for designers and developers. Markie collects images, videos, PDFs, SVGs, and website bookmarks via a Chrome extension, then enriches them with AI-generated descriptions, tags, and color palettes. This MCP server lets AI assistants search, browse, and retrieve assets from your Markie library.

Prerequisites

1. The Markie desktop app must be running on your machine (it serves the API on localhost:3210).
2. Node.js 18 or later.

Setup

No API key required — the server authenticates automatically with the local Markie app.

Claude Code

claude mcp add markie -- npx -y markie-mcp

Claude Desktop

Add to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json):

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

Cursor / other MCP clients

Use npx -y markie-mcp as the command. The server communicates over stdio.

Available Tools

search_assets

Full-text search across your library with optional filters.

| Parameter | Type | Required | Description | |-----------|------|----------|-------------| | query | string | No | Search query (searches descriptions, titles, tags, URLs, categories) | | tags | string[] | No | Only assets that have all of these tags | | type | string | No | Asset type: image, video, pdf, vector, or website | | dateFrom | string | No | ISO date — assets saved on or after this date | | dateTo | string | No | ISO date — assets saved on or before this date | | colors | string[] | No | Hex colors — assets with at least one matching palette color | | limit | number | No | Max results to return (default 20, max 100) |

get_asset_detail

Full metadata for a single asset.

| Parameter | Type | Required | Description | |-----------|------|----------|-------------| | id | string | Yes | Asset ID |

get_asset_thumbnail

Thumbnail image for an asset. Returns a base64-encoded JPEG so the AI can see the asset visually.

| Parameter | Type | Required | Description | |-----------|------|----------|-------------| | id | string | Yes | Asset ID |

get_asset_file

Full file content as base64. Returns an error if the file exceeds 10 MB.

| Parameter | Type | Required | Description | |-----------|------|----------|-------------| | id | string | Yes | Asset ID |

list_tags

All tags in your library, sorted by frequency (most used first). No parameters.

list_colors

All hex colors extracted from asset color palettes, sorted by frequency. No parameters.

get_recent

Most recently saved assets.

| Parameter | Type | Required | Description | |-----------|------|----------|-------------| | n | number | No | Number of assets to return (default 10, max 50) |

get_stats

Library statistics: total asset count, breakdown by type, total storage size, unique tag count, AI-enriched count, and date range. No parameters.

Asset Schema

Each asset returned by the tools contains these fields:

| Field | Type | Description | |-------|------|-------------| | id | string | Unique identifier | | sourceUrl | string | Original URL the asset was saved from | | sourceDomain | string | Domain of the source URL | | pageTitle | string | Title of the page the asset was found on | | pageUrl | string | Full URL of the source page | | type | string | image | video | pdf | vector | website | | mimeType | string | MIME type (e.g. image/png) | | fileName | string | Stored file name | | originalName | string | Original file name | | fileSize | number | File size in bytes | | width / height | number | null | Dimensions (images and videos) | | savedAt | string | ISO 8601 timestamp | | tags | string[] | AI-generated and manual tags | | description | string | null | AI-generated description | | colorPalette | string[] | null | Extracted hex color palette | | category | string | null | AI-assigned category | | aiEnriched | boolean | Whether AI enrichment has been applied |

How It Works

The MCP server connects to the Markie desktop app's local API on localhost:3210. It registers for a bearer token on startup and caches the asset catalog in memory with a 5-second TTL. Full-text search uses fuzzy matching across descriptions, titles, tags, URLs, and categories.

If Markie isn't running when a tool is called, you'll get a friendly error message asking you to start the app. The server will re-register automatically on the next call once Markie is back.

Troubleshooting

"Markie is not running" — Start the Markie desktop app. The MCP server needs it running on localhost:3210.

Assets not showing up — The server caches the catalog for 5 seconds. Newly saved assets will appear after a short delay.

Tools work at first, then stop — If you quit Markie after the MCP server started, tools will return a connection error. Restart Markie and the server will reconnect automatically.

License

MIT