The Goal
By the end of this blueprint, you will have a working MCP server that:
- Saves bookmarks with titles, URLs, tags, and notes
- Searches your collection by keyword, tag, or domain
- Organizes with tags — add, remove, and filter by tags
- Exposes your collection as resources Claude can browse
- Runs locally and connects to Claude Desktop or Claude Code
You will open Claude and say things like:
"Save this article: https://example.com/great-post — tag it with 'AI' and 'research'"
"What bookmarks do I have about machine learning?"
"Show me everything I saved this week"
...and Claude will execute these against your local bookmark database.
Why This Stack
| Choice | Why |
|---|---|
| Node.js + TypeScript | The MCP SDK is most mature in TypeScript. Most developers already know JS. |
| STDIO transport | The simplest transport — no HTTP server, no ports, no auth. Claude Desktop launches your server as a subprocess. |
| SQLite (via better-sqlite3) | Zero-config database. No external services. Your data stays on your machine. |
| Official MCP SDK | @modelcontextprotocol/sdk — the reference implementation maintained by Anthropic. |
How MCP Works (30-Second Version)
MCP defines three primitives your server can expose:
- Tools — Actions Claude can take (save a bookmark, search, tag)
- Resources — Data Claude can read (your bookmark collection, tag list)
- Prompts — Reusable templates (weekly digest, tag cleanup)
Claude discovers these capabilities automatically when it connects to your server.
Project Structure
Here is what we will build:
bookmark-mcp/
├── src/
│ ├── index.ts # Server entry point
│ ├── db.ts # SQLite database setup and queries
│ ├── tools.ts # MCP tool definitions
│ ├── resources.ts # MCP resource definitions
│ └── prompts.ts # MCP prompt templates
├── package.json
├── tsconfig.json
└── bookmarks.db # Created automatically at runtimeLet's start building.