Add Bookmarks to Claude in 5 Minutes (MCP Setup)

Here is what the experience looks like once your bookmarks are connected.

You are deep in a debugging session. Halfway through, you remember bookmarking a thread about JavaScript event loop quirks six months ago. Instead of opening X, scrolling through hundreds of saved tweets, and hoping you can find it, you just ask Claude: “Do I have anything saved about the JavaScript event loop?”

Claude calls search_bookmarks("JavaScript event loop") and returns three relevant tweets with full content. You get the answer without leaving your terminal.

That is the whole pitch. If it sounds useful, here is how to set it up.

What is the Model Context Protocol and why does it help?

The Model Context Protocol (MCP) is an open standard developed by Anthropic that lets AI tools connect to external data sources. Instead of copying and pasting content into a chat window, the AI can query live data directly during a conversation.

For a full breakdown of how MCP works and why it matters, read What Is MCP? The Protocol Connecting AI to Your Data. You can also check our MCP glossary entry for a concise definition. The short version: it turns Claude from a tool you talk to into a tool that can reach into your other systems. ContextBolt also has a directory entry in the MCP server catalog with example prompts and one-click install snippets. If memory is the broader story you are after, see 9 ways to give Claude long-term memory.

ContextBolt exposes six MCP tools over your bookmark collection: semantic search, topic cluster listing, cluster browsing, recent saves by platform, exporting the collection, and saving a new bookmark. When you ask Claude a question that touches on your saved content, it calls the right tool automatically. You do not need to learn any commands or mention tools by name.

This is significant for one reason that is easy to underestimate. Most developers have been building a personal knowledge base for years without realizing it. Every tweet you bookmarked about system design, every Reddit thread you saved about hiring, every LinkedIn post you flagged about pricing strategy: that is a curated body of knowledge. It just has no interface. MCP gives it one.

What you need before connecting bookmarks to Claude Code

Three things:

1. ContextBolt installed: The free Chrome extension captures your X/Twitter, Reddit, and LinkedIn bookmarks automatically. Start at contextbolt.com.
2. A Pro subscription: The MCP endpoint is Pro-only ($6/month). Free users get AI tagging, topic clustering and instant local search inside the extension. Semantic search runs server-side, so it arrives with Pro’s MCP endpoint. Pro also includes unlimited bookmarks and cloud sync.
3. Claude Code, Cursor, or Windsurf: Any AI coding tool that supports remote MCP servers via HTTP transport works. The setup differs slightly for each one.

One other prerequisite worth flagging. The MCP endpoint queries your bookmark data from the cloud, so nothing works until the first sync has run. There is no switch for it: sync starts on its own the moment your Pro license is active, and runs in the background from then on. If the server says no data is available, give that first upload a minute.

Step 1: Use the Bookmarks server URL

Use https://api.contextbolt.app/mcp as the Bookmarks MCP URL. It contains no personal key. Your Pro account supplies access when you sign in.

Each app needs its own connection. Use the email on your ContextBolt Pro account and complete the email-code sign-in when prompted.

Step 2: Add to Claude Code

Claude Code has a CLI command that handles MCP server registration. Run this in your terminal.

claude mcp add --transport http bookmarks https://api.contextbolt.app/mcp

By default, this stores the server at --scope local, available in your current project. To make it available across every project, add --scope user:

claude mcp add --transport http --scope user bookmarks https://api.contextbolt.app/mcp

Open Claude Code and run /mcp. Select bookmarks and authenticate. Enter your Pro account email and email code in the browser, then allow the connection. After signing in, verify it worked.

claude mcp list

You should see bookmarks listed with a connected status.

Alternative: edit .mcp.json directly

If you prefer manual config (or want to commit a shared setup to a repo), add this to .mcp.json in your project root:

{
  "mcpServers": {
    "bookmarks": {
      "type": "http",
      "url": "https://api.contextbolt.app/mcp"
    }
  }
}

Note the "type": "http" field. Claude Code requires this for remote HTTP servers. Omitting it will cause the connection to fail.

The shared URL is safe to commit. Each user authenticates separately through /mcp. Keep email codes, access tokens, and any older personal URLs out of the repository.

Step 3: Add to Cursor

Open (or create) .cursor/mcp.json in your project root, or the global version at ~/.cursor/mcp.json. Add:

{
  "mcpServers": {
    "bookmarks": {
      "url": "https://api.contextbolt.app/mcp"
    }
  }
}

Cursor uses "url" directly with no "type" field needed. Save the file and reload the Cursor window. Authenticate the connection before using the bookmark tools in Cursor chat.

In Cursor’s MCP settings, use the connection’s sign-in option and complete the ContextBolt email-code flow with your Pro account. Saving the JSON alone does not authenticate.

Step 4: Add to Windsurf

Windsurf uses a different config file and a different field name for remote URLs. Open:

• macOS / Linux: ~/.codeium/windsurf/mcp_config.json
• Windows: %USERPROFILE%\.codeium\windsurf\mcp_config.json

If your Windsurf version supports remote MCP with OAuth, add this under mcpServers.

{
  "mcpServers": {
    "bookmarks": {
      "serverUrl": "https://api.contextbolt.app/mcp"
    }
  }
}

The key difference: Windsurf uses "serverUrl" not "url". Using "url" will silently fail. Save the file, refresh MCP servers, and complete the connection’s browser sign-in with your Pro account. If your version cannot complete OAuth, use the personal URL from Settings → MCP Integration → Older clients in the extension instead. That URL contains a secret and must stay private. Full details at Windsurf’s MCP documentation.

What about connecting bookmarks to Claude Desktop?

The Claude Desktop app (the mac and windows GUI client) does not support remote MCP servers via the claude_desktop_config.json file. That config file only works for local stdio servers that run as a process on your machine.

To connect ContextBolt to Claude Desktop, use the Connectors UI instead. Go to claude.ai/settings, find Connectors, and paste your full ContextBolt MCP endpoint URL (https://api.contextbolt.app/mcp). Click Add, enter your Pro account email and email code, and allow the connection. Connections made this way are brokered through Anthropic’s infrastructure rather than running locally on your device.

The five ContextBolt MCP tools and what to ask

Once connected, Claude has access to six tools from your bookmark collection. It chooses which to call based on what you ask. You never need to invoke them manually.

Search is semantic, not keyword-based. If you ask “what do I know about managing a small team?” and your bookmarks contain threads about “headcount planning” and “flat org structures”, they will surface. The endpoint embeds your query with the same model used on your bookmarks at save time and compares the two. That embedding search is server-side, which is why it lives on the MCP endpoint rather than inside the extension, where search is a local ranked index.

You can also filter by platform. “What have I saved about hiring on LinkedIn?” limits results to LinkedIn saves. “Show me my recent Reddit bookmarks” returns only Reddit. The platform parameter accepts twitter, reddit, linkedin, or all.

Real bookmark MCP use cases worth trying

A few scenarios where this pays off immediately:

Researching a technical decision: This is where developers get the most value. You are choosing between two approaches and want to know what you have read about them. “Do I have bookmarks on database connection pooling?” surfaces relevant threads you saved months ago and had forgotten about.

Preparing for a conversation: You are about to talk to someone about a topic and want to recall what you know. “What do I have saved about B2B pricing models?” pulls together your saved content before you get on the call.

Recovering a specific source: You remember saving a tweet from someone you follow but cannot recall the details. “Find the bookmark from @something about database indexing” retrieves it with the full text, even if the original tweet has since been deleted from X.

Exploring what you know: “List my topic clusters” gives you a map of your own saved knowledge. It is often surprising how much you have accumulated on a topic without realizing it.

That last point is the one worth sitting with. Most developers I know have been bookmarking content for years. The average active user saves somewhere between 200 and 500 items per year. Across three or four years, that is a serious body of curated material. The problem was never a lack of content. It was a lack of interface.

A comparison of MCP setup methods for Claude Code

The most common mistake is using the wrong field name. Claude Code and Cursor use "url". Windsurf uses "serverUrl". Claude Code also requires "type": "http" which Cursor does not. Get these wrong and the connection will fail silently, which is frustrating to debug.

Why bookmarks in Claude Code matters more than it first appears

The obvious use case is “search your bookmarks without switching apps.” That is genuinely useful, but it undersells what is actually happening.

When your bookmarks are connected to Claude, your AI assistant has context you have been building for years. Every thread you saved because it was interesting, every post you bookmarked because you knew you would need it later: all of it becomes available in every conversation, without you having to remember it or go looking.

This is the version of a personal knowledge base that most people have been trying to build with Notion databases and elaborate folder systems. The difference is that this one runs without any maintenance. You save things as normal. The extension handles the AI tagging and cloud sync. The MCP endpoint makes it all available inside your tools.

An MCP endpoint is no longer rare. Readwise ships an official one, Raindrop ships one in beta for Pro users, and community servers cover most of the rest, so the Model Context Protocol is a checkbox rather than a moat. What an agent gets through any of them is whatever the library already holds. ContextBolt’s is the only one behind automatic X, Reddit and LinkedIn capture, keeping the full post rather than a link that rots, which is why the answers coming back are different even when the plumbing is the same.

ContextBolt is a Chrome extension that captures your X, Reddit, and LinkedIn bookmarks into an AI-powered knowledge base. The Basic tier is free and includes 150 bookmarks, AI tagging, topic clustering, and instant local search. Pro ($6/month) adds unlimited bookmarks, cloud sync, and the MCP endpoint covered in this guide.