LinksBubble just got smarter. Connect your personal reference archive to Claude, Cursor, Windsurf, or VS Code instantly using the Model Context Protocol (MCP). Zero cloud, 100% private.
Bridging the gap
You've built a curated archive of references, files, and annotations in LinksBubble. Now, your AI assistant has native access to it.
Every time you asked Claude or Cursor a question, it lacked context. You had to copy-paste bookmarks, re-explain documentation, or completely forget that you already saved the perfect reference days ago. The gap between your saved wisdom and your AI was slow, manual, and frustrating.
With built-in MCP protocol support, LinksBubble acts as a local context server. When you ask your IDE or AI tool a question, the assistant instantly reaches out to LinksBubble locally, parses your titles, tags, and personalized notes, and gives you hyper-focused answers tailored directly to what you know.
Smarter Workflows
Because the AI reads your tags and personal notes alongside URLs, it understands your developer intent, not just raw links.
Quick Integration
No packages to build. LinksBubble bundles the compiled `linksbubble-mcp.exe` server natively. Simply select your tool below and connect.
Open LinksBubble, go to the ••• Menu -> MCP Integration, and click Connect Claude. Or, edit your manual configuration file directly at:
If configuring manually, insert the following entry inside the "mcpServers" block (replacing YourName with your actual Windows username):
{
"mcpServers": {
"linksbubble": {
"command": "C:\\\\Users\\\\YourName\\\\AppData\\\\Local\\\\Programs\\\\linksbubble-release\\\\resources\\\\linksbubble-mcp.exe"
}
}
}
Relaunch your Claude Desktop app. Claude will instantly connect to LinksBubble and read all your saved reference points locally!
Claude App Setup Walkthrough
Open LinksBubble's ••• Menu -> MCP Integration and click the copy icon next to the server path, or copy the standard path below:
C:\\Users\\YourName\\AppData\\Local\\Programs\\linksbubble-release\\resources\\linksbubble-mcp.exe
Open Cursor and go to Settings -> Cursor Settings -> Features -> MCP, or configure it manually inside the global storage configuration file:
Add the `linksbubble` entry in the `mcpServers` object and replace `YourName` with your Windows username:
{
"mcpServers": {
"linksbubble": {
"command": "C:\\Users\\YourName\\AppData\\Local\\Programs\\linksbubble-release\\resources\\linksbubble-mcp.exe"
}
}
}
Save the configuration and restart Cursor. You will see a green status dot in Cursor Settings showing the LinksBubble MCP is active and connected!
Cursor IDE Setup Walkthrough
Open Windsurf settings or directly open your global Windsurf MCP configuration file:
Add the following configuration block (ensure to replace `YourName` with your actual Windows username):
{
"mcpServers": {
"linksbubble": {
"command": "C:\\Users\\YourName\\AppData\\Local\\Programs\\linksbubble-release\\resources\\linksbubble-mcp.exe"
}
}
}
Restart Windsurf. The assistant is now fully empowered to query your local bookmark databases.
Windsurf Setup Walkthrough
Open VS Code, press `Ctrl+Shift+P`, type `Continue: Open Config`, or edit the file directly at:
Add the following entry under the `mcpServers` array in the JSON file:
{
"name": "linksbubble",
"command": "C:\\Users\\YourName\\AppData\\Local\\Programs\\linksbubble-release\\resources\\linksbubble-mcp.exe"
}
Press `Ctrl+Shift+P` in VS Code and select `Developer: Reload Window` to connect Continue to your LinksBubble MCP server.
Continue (VS Code) Setup Walkthrough
Since ChatGPT Web is cloud-hosted, it requires a secure public HTTPS endpoint rather than localhost. Expose your local LinksBubble MCP server port using ngrok or a similar tunneling tool:
Go to ChatGPT, click your profile menu, open Settings → Apps → Advanced settings and toggle Developer mode to ON.
Click Add MCP Server, choose SSE transport, and paste your public tunnel URL (e.g. https://your-tunnel.ngrok-free.app/sse):
Server Type: SSE URL: https://your-tunnel.ngrok-free.app/sse
ChatGPT Setup Walkthrough
Transparent Integration
We believe in ultimate data clarity. Only clean metadata required for context is shared with the AI model. No visual bulk, layout data, or thumbnails leave the server.
| Field | Data Shared with AI | Purpose |
|---|---|---|
| URL | The web address or local document path (e.g. github.com/... or C:/Docs/...) |
Allows the AI to point you directly to the correct reference document or site. |
| Title | The retrieved page title or document name (e.g. React Boilerplate Template) |
Helps the AI identify and search for references based on titles. |
| Note | Your custom text comments and annotations added when saving the link | Critical developer context. Tells the AI your specific intent and thoughts. |
| Tags | The array of tags you assigned (e.g. #auth, #react, #api) |
Enables targeted filtering so the AI can group relevant resources. |
| Saved on | Timestamp of when you pinned the reference | Allows queries based on time (e.g. "What did I save yesterday?"). |
The LinksBubble MCP server is a local background utility that runs entirely on your system. It reads your local links.json and communicates directly with your local IDE (Cursor/Windsurf) or Claude Desktop using secure stdin/stdout pipes. Nothing is uploaded to our servers, nothing leaves your computer, and no subscription cloud is required. It is context management the way it should be: fully private and under your control.