Bookmarks MCP Server: Search Your Saves in Claude & Cursor

The Bookmarkjar bookmarks MCP server lets your editor's AI reach your saved links without interrupting your work. Ask Claude Code to find your Drizzle notes, save a useful URL, or prepare a duplicate review, and it can call Bookmarkjar directly.

Bookmarkjar exposes a remote Model Context Protocol server using Streamable HTTP and OAuth. You authorize it in the browser, so there is no API key to paste into an editor configuration.

What the Bookmarkjar MCP server can do

The server exposes three tools:

• bookmarks.search searches your bookmarks by text, tags, domain, type, or visibility.
• bookmarks.insert saves one or more URLs to your library.
• cleanup.create creates a review of duplicate or unwanted bookmarks without deleting anything automatically.

Read access and write access use separate OAuth scopes. Your MCP client requests the permissions it needs and Bookmarkjar shows them on the consent screen before issuing a token.

Connect Bookmarkjar to Claude Code

Claude Code supports remote HTTP MCP servers. Add Bookmarkjar with one command:

claude mcp add --transport http bookmarkjar https://bookmarkjar.com/api/mcp

Claude Code should open Bookmarkjar in your browser. Sign in, review the requested search and write permissions, and choose Allow. The client then completes the OAuth authorization-code flow with PKCE and discovers the available tools.

The Claude Code MCP documentation explains how to inspect and manage configured servers.

Connect Bookmarkjar to Cursor or another MCP client

Use the client's remote MCP setup flow with this endpoint:

https://bookmarkjar.com/api/mcp

For clients that accept JSON configuration, the server entry is typically:

{
  "mcpServers": {
    "bookmarkjar": {
      "type": "http",
      "url": "https://bookmarkjar.com/api/mcp"
    }
  }
}

Do not add an x-api-key header. A compatible client discovers Bookmarkjar's OAuth authorization server, opens the browser consent flow, and stores the resulting scoped token.

Search bookmarks from your editor

Ask naturally:

• "Search my bookmarks for Postgres row-level security."
• "Find the design references I saved from example.com."
• "Show my bookmarks tagged with TypeScript."

The bookmarks.search tool is always restricted to the organization selected for the signed-in OAuth user. Search calls use Bookmarkjar's normal indexed search path, so the results match the library rather than a separate MCP-only store.

Save links and create cleanup reviews

The write scope enables bookmarks.insert and cleanup.create. For example:

• "Save https://orm.drizzle.team/docs/overview to my bookmarks."
• "Create a cleanup review for duplicate GitHub bookmarks."

Inserted URLs go through Bookmarkjar's regular bookmark creation and ingest pipeline. Cleanup requests use the same premium checks, daily accounting, queue dispatch, and review UI as cleanup requests created inside the app.

A cleanup tool call only creates a review. You still choose what to keep or delete in Bookmarkjar before any destructive cleanup action runs.

How OAuth keeps access scoped

Bookmarkjar publishes standard OAuth authorization-server and protected-resource metadata for the MCP endpoint. Dynamic client registration and PKCE let compatible clients connect without a pre-shared client secret.

Tokens are issued for the /api/mcp audience and can include mcp:read, mcp:write, or both. Bookmarkjar derives the user and organization from the validated token rather than accepting ownership identifiers from tool arguments.

Troubleshooting the connection

If the consent page does not open, confirm that your client supports remote Streamable HTTP MCP with OAuth discovery. Remove and re-add the server if the client cached an earlier registration or authorization request.

If a tool is missing, reconnect and approve the scope it requires. Search needs mcp:read; insert and cleanup need mcp:write. An active paid Bookmarkjar plan is required for MCP access.

Once connected, your saved research is available where you are already working—without copying credentials into config files or switching back to the browser for every search.