[Feature Request] Event-driven co-browsing mode for Claude in Chrome extension

Resolved 💬 1 comment Opened Apr 19, 2026 by akumeitor-art Closed May 27, 2026

Preflight Checklist

  • [x] I have searched existing requests and this feature hasn't been requested yet
  • [x] This is a single feature request (not multiple features)

Problem Statement

Currently, having Claude observe a user's browsing session requires either manual screenshot requests ("take a screenshot now") or continuous high-frequency polling. Both approaches are impractical:

  • Manual requests break the natural flow of browsing and require constant user intervention
  • - Continuous polling consumes a disproportionate amount of tokens and context window, making real-time co-browsing unsustainable

There is no way for Claude to naturally "accompany" a user while they browse without one of these two costly trade-offs.

Proposed Solution

Add an optional "co-pilot" mode to the Claude in Chrome extension that triggers inference based on configurable browser events, rather than requiring manual screenshots or continuous polling.

The extension would monitor configurable browser events and send Claude a snapshot (screenshot + context) only when they occur:

  • Navigation to a new URL
  • - Form submission — Claude sees the filled form and the result after clicking Accept/Submit
  • - - Right-click context menu — the option selected and its result (e.g. "Save as...")
  • - - - File download initiated — filename and source URL
  • - - - - Any click producing a page change or browser dialog

Configurability options:

  • Toggle the mode on/off
  • - Set event grouping threshold (e.g. notify every 1, 3 or 5 events instead of every single one)
  • - - Choose which event types to monitor

This approach keeps inference contained and predictable — Claude only processes when something relevant happens, not continuously.

Alternative Solutions

  • Manual screenshots: works but breaks browsing flow and requires constant user intervention
  • - Continuous polling: technically possible but consumes too many tokens and fills the context window rapidly, making it unsustainable for any session longer than a few minutes
  • - - Windows Problem Steps Recorder (psr.exe): captures steps automatically but saves everything at the end as a static file, with no real-time integration with Claude

Priority

Medium - Would be very helpful

Feature Category

Other

Use Case Example

Scenario: A user wants Claude to assist them while filling out their annual tax declaration online.

  1. User enables "co-pilot" mode in the Claude in Chrome extension
  2. 2. User navigates to the tax agency website — Claude receives a snapshot and sees the page
  3. 3. User fills in fields and clicks "Next" — Claude receives the form data and the result page
  4. 4. Claude proactively suggests corrections or flags potential issues at each step
  5. 5. User downloads a PDF confirmation — Claude is notified of the download and can reference it

Without this feature, the user would need to manually tell Claude "take a screenshot now" at every single step, completely breaking their focus and flow.

Additional Context

_No response_

View original on GitHub ↗

This issue has 1 comment on GitHub. Read the full discussion on GitHub ↗