Free · MIT · Runs locally · No account

The bookmarks you already have deserve a decision

You have a few thousand bookmarks scattered across browsers and read-later exports. Not one of them has ever been decided about — they were saved, and then they were forgotten. These three small tools fix that in order: read them all, let your AI assistant search them, then decide about a handful every day.

1
parse-bookmarksFor developers

Read them all — every export, one format

Chrome, Firefox, Safari and Edge export a 1996-era HTML file with unclosed tags. Pocket, Instapaper and Raindrop each export a different CSV. This zero-dependency library reads all of them into one clean list — title, URL, date saved, tags, folder — and converts between formats.

It also knows about the Chrome file most tools miss: signed-in Chrome keeps synced bookmarks in AccountBookmarks, not Bookmarks. Tools that only open the second file show you an empty list.

npm install parse-bookmarks
2
bookmarks-mcpFor anyone using Claude or Cursor

Let your AI assistant see them

A read-only MCP server. Install it and Claude, Cursor or Claude Code can search the bookmarks sitting in Chrome, Edge, Brave, Arc, Vivaldi, Firefox and Safari (Reading List included) — plus any export file. Ask “what did I save about pricing pages?” and get an answer from your own saves.

Read-only by design: it cannot write to your browser, phones home to nothing, and needs no account or API key.

claude mcp add bookmarks -- npx -y bookmarks-mcp
3
bookmarks-lifecycleFor anyone with a backlog

Decide about a few every day

Seeing 3,000 bookmarks is not the same as dealing with them. This server turns the pile into a daily to-do list: 15 bookmarks a day come up for a decision, and you have 24 hours to keep, watch for 30 more days, or let go. Undecided ones lapse on their own.

Nothing is ever deleted. A lapsed bookmark leaves your attention, not your disk, and every change — including the ones an AI made — can be undone.

claude mcp add bookmarks-lifecycle -- npx -y bookmarks-lifecycle

What all three promise

  • Everything runs on your machine. No account, no token, no API key.
  • Nothing is sent anywhere. Search the source for “fetch” — there is none.
  • Your browser is never written to. The only file that changes is a local state file you can delete.
  • MIT licensed. Fork it, vendor it, ship it inside your own thing.

The three are independent. Use the parser alone in your own project, the MCP server alone for search, or all three together — the lifecycle server reads through the MCP server, which reads through the parser.

And the bookmarks you save from now on?

These tools clean up the past, one decision at a time, whenever you ask. Burn 451 is the version that runs by itself for everything new: every link you save gets a 24-hour countdown — read it, keep it, or let it burn.

The reading flow is free. AI summaries, full-text search and unlimited Vault are $4.99/mo.