Reading Queue MCP

Pocket inside Claude — paste a URL in chat, Claude saves it to a SQLite-backed reading queue, and a Prefab dashboard shows what you have to read.

Tools

Tool	Kind	What it does
fetch_url_metadata(url)	internet	Fetches the page, parses <title> / <meta description> / OpenGraph tags.
manage_reading_list(action, ...)	local CRUD	SQLite-backed. Actions: add, list, mark_read, delete.
render_reading_queue()	UI	Returns a Prefab dashboard showing unread queue + recently read items + counts.

Plus an MCP prompt template save_and_show(url) that calls all three in one go.

Killer prompt

Save https://example.com/some-article to my reading list and show me my queue.

The agent must hit all three tools to answer.

Run locally

uv sync
uv run reading-queue-mcp

Test interactively with the MCP inspector:

uv run mcp dev src/reading_queue_mcp/server.py

Wire into Claude Desktop

Edit ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "reading-queue": {
      "command": "uv",
      "args": ["--directory", "/absolute/path/to/reading-queue-mcp", "run", "reading-queue-mcp"]
    }
  }
}

Restart Claude Desktop.

Optional env

• READING_QUEUE_DB — overrides the SQLite path (default ~/.reading_queue.db). Handy for tests / demos.