How it works

Agentic Acceleration

AI coding agents are first-class users of Agentic Bookmarks, not an afterthought. The bundled MCP server gives them the same view a human has — plus the tools they're uniquely good at using.

Video coming soon

Bookmarks as task context

When an agent picks up a ticket, the first thing it needs is a map of the relevant code. Bookmarks — with their labels and notes — are exactly that map, written by the people who already understand the system. The agent pulls the bookmarks that touch the area it's working in and feeds them into its working context, instead of re-deriving the lay of the land from a cold read of the repo every time.

The bookmark set becomes an evolving project memory: a curated, durable, queryable substrate that both humans and agents contribute to and draw from.

Research and placement

Agents are well-suited to investigation work — tracing a callsite, mapping a subsystem, finding every place a flag is read. At the end of that work, instead of dumping their findings into a chat that scrolls away, agents can place bookmarks with strong labels and notes at the right lines. The investigation persists, attached to the code, useful to the next agent and the next teammate who lands in the same area.

A shared interface for review and curation

Bookmarks turn out to be a surprisingly good interface between humans and agents. An agent can stage a set of bookmarks at the lines it wants the human to review; the human can drop in, walk the pinned set, and respond directly — accept, edit, regroup, or retire. The conversation lives next to the code instead of in a chat transcript that loses its referent the next time the file changes.

Agents can also rename, retag, and clean up bookmarks as part of broader maintenance — the same hygiene work a careful teammate would do, applied at machine pace.

Repair as an agentic strength

The hardest part of making bookmarks worth checking in is what happens when an anchor breaks. Mechanical auto-repair handles most drift, but some cases need judgment — code that was moved across files, refactored into a different shape, or rewritten outright. The MCP repair surface gives agents a structured waterfall — diff diagnosis, baseline-commit forensics, line tracing, moved-code search, cross-file search — so the rare hard case gets resolved instead of leaving a silent broken pin behind. This is the capability that makes long-lived shared bookmarks practical.

Bundled, no separate setup

The MCP server ships with the extension. Point a compatible agent at it and the full surface — read, place, organize, repair — is available immediately. No second install, no separate process to manage.

← How anchors work← Local vs. shared bookmarks