BookmarkOps

Sync-safe AI bookmark cleanup. Review first, apply safely.

Your bookmarks are probably a mess. Years of "I'll get to that later" piled into folders nobody opens. You've thought about cleaning them up, but the prospect of manually re-sorting a thousand entries is exactly why it never happens. BookmarkOps lets an AI tool — OpenAI Codex, Claude Code, or Cursor — do the planning work for you. You ask it to scan your bookmarks and propose a cleanup plan. BookmarkOps generates a structured report, the AI drafts the plan, and the extension queues that plan for your review. Nothing in your bookmark tree changes until you type the word APPLY in the dashboard. What is MCP, and why does it matter? MCP — the Model Context Protocol — is a small bridge that lets AI tools talk to local applications safely. With BookmarkOps's MCP server connected, your AI tool can ask the extension to: - Scan your bookmark tree and return a report. - Submit a cleanup plan for your review. - Check whether the plan would succeed (a "dry run"). What the AI tool cannot do is change anything. It can only propose. You're the only one who can hit APPLY. Three things that make BookmarkOps different 1. Runs entirely on your machine. Your bookmarks are not uploaded anywhere. The MCP bridge binds to localhost:7842 and is unreachable from outside your computer. No telemetry, no analytics, no remote logging, no third-party SDK. Every byte of code is open source and auditable. 2. Three-layer confirmation before any destructive change. Applying a plan, restoring a backup, and deleting a backup each require typing a different confirmation phrase — APPLY, RESTORE, DELETE BACKUP. Before any apply, a full snapshot of your bookmark tree is taken automatically. You can roll back to any prior snapshot with a single typed confirmation. 3. Open source under MIT. Every line of the BookmarkOps extension and the @bookmarkops/mcp Node package is on GitHub. You can read the code, build it yourself, fork it, audit it. There is no closed-source server, no proprietary backend, no future pivot into telemetry. How the workflow runs scan -> report -> plan -> dryRun -> preview -> backup -> apply -> verify -> restore 1. Scan — BookmarkOps reads your bookmark tree through Chrome's official chrome.bookmarks API. 2. Report — An AI-readable Markdown or JSON summary is generated locally. 3. Plan — Your AI tool drafts a bookmark-plan.json and submits it through MCP. 4. DryRun — BookmarkOps simulates the plan against a copy of your tree and shows you exactly what would change. 5. Preview — Grouped, color-coded operations appear in the dashboard. 6. Backup — Before any change runs, the full bookmark tree is snapshotted. 7. Apply — Only after you type APPLY do the changes happen. 8. Verify — BookmarkOps re-checks the live tree against the plan and reports what passed or failed. 9. Restore — If anything looks wrong, restore from any prior backup with a typed RESTORE. Quick start First-time users often find OpenAI Codex the easiest start — its interface is a familiar AI chat, similar to ChatGPT, Claude, or Gemini. BookmarkOps works identically with Claude Code or Cursor; pick whichever AI tool you already use. 1. Install BookmarkOps from the Chrome Web Store. 2. Open the dashboard, copy the MCP configuration block from the Quick Setup card. 3. Paste it into your AI tool (Codex, Claude Code, or Cursor). 4. Ask the AI: "Scan my bookmarks and suggest a cleanup plan." 5. Review the proposed changes in the dashboard. Approve or reject. Who BookmarkOps is for - People who already work with AI tools. If you already use Codex, Claude Code, or Cursor, BookmarkOps adds bookmark cleanup to the same conversation. No extra account, no separate tool to learn. - People with thousands of unorganized bookmarks. If your bookmark bar has been growing for years, the AI can do the tedious classification work in minutes. - Privacy-conscious users. If you don't want your bookmark titles uploaded to a cloud service for "AI organization", BookmarkOps keeps everything local. Frequently asked questions Do I need to install anything besides the extension? You need an AI tool that supports MCP (Codex, Claude Code, Cursor, or anything else that speaks the protocol). The MCP bridge (@bookmarkops/mcp) is fetched automatically by npx the first time your AI tool starts it — no manual installation. Can I use BookmarkOps without an MCP-capable AI tool? Not in this release. The cleanup workflow currently requires an MCP-capable AI tool to draft and submit the plan. The shortest path is OpenAI Codex (see Quick start above). A file-based manual mode — drop in a plan JSON without any AI tool — is on the roadmap for a future release. What happens if a cleanup goes wrong? Every apply takes a backup first. Type RESTORE in the dashboard and the previous bookmark state is restored, exactly as it was. What if I rotate my MCP token? Rotate in the dashboard, copy the new Quick Setup config, then restart your AI tool. The old token stops working once your AI tool restarts. See the agent documentation in the repository for the full rotation flow and security model. Open source BookmarkOps is open source under the MIT License. The Chrome extension and the @bookmarkops/mcp Node package both live in the same GitHub repository, with the privacy policy, threat model, and architecture documentation alongside the source code. GitHub: https://github.com/brianjhang/bookmarkops