BrowserAgent

Browser context for your coding agent

Screenshots, elements, and errors delivered to Claude Code, Codex CLI, and Gemini CLI in one click.

Agent Logger
Online
Claude Code
terminal_vision
Connected to
terminal_vision
Claude Code MCP
Native MCP
Errors0
AllErrorsWarningsNetwork
Search errors
No errors detected

Works with

Claude CodeCodex CLIGemini CLIOther MCP clients

Capture and send in one click

Viewport, full-page, region, or element. Annotate, then send the image to the agent in the terminal.

  • 01Viewport, full page, region, or a single element
  • 02Annotate with arrows, text, and highlights
  • 03Send to the paired session, or save locally
Screenshot preview

1421 x 1042 - viewport

SaveSend

Select any element

Click a node on the page. The side panel collects selector, dimensions, computed styles, and accessibility data, then sends it to the agent.

Selected element

selector
#submit-btn
tag
button
text
Submit
role
button
size
366 × 208
Copy selectorHighlightSend

Catch errors as they happen

Console errors, network failures, and uncaught exceptions are captured in the side panel, ready to send to the agent.

Errors2
  • TypeError

    Cannot read properties of undefined

    src/api/users.ts:47

  • Failed fetch

    GET /api/session 500

    network

Errors include the exact source

File path, line number, function name, and stack trace go with the error. The agent can open the file instead of searching the repo.

Without source context

TypeError: Cannot read properties of undefined No source information available
  • Search the codebase for a matching string
  • Read multiple files to find the source
  • Extra round trips and wasted tokens

With Agent Logger

TypeError: Cannot read properties of undefined at fetchUserData (src/api/users.ts:47:12) at Dashboard (src/components/Dashboard.tsx:23:5)
  • Exact file and line number
  • Full stack trace with calling functions
  • The agent opens the failing line

MCP-native browser control

Built on the Model Context Protocol. The terminal agent can request screenshots, click elements, and automate the page through conversation.

Browser

Side panel

Capture, select, stream errors

Agent

Claude Code, Codex CLI, Gemini CLI

Manual configuration is available for other MCP-compatible clients.

  • browser_screenshot

    Capture screenshots

    Request viewport, full-page, or element screenshots from the terminal.

  • browser_click

    Click and type

    Interact with page elements through conversation. Click, type, and fill forms.

  • browser_get_errors

    Get errors

    Fetch console errors with stack traces, file path, and line number on demand.

  • browser_scroll

    Automate the page

    Scroll, hover, and navigate without leaving the terminal.

What the agent can see

Selectors, screenshots, errors, and page context, sent from the side panel to your terminal.

Element Selection

Click any element on the page. Get selector, dimensions, computed styles, and accessibility info, all sent to your AI assistant in one click.

Screenshot Capture

Four capture modes: viewport, full-page, region, or element. Annotate screenshots with arrows, text, and highlights before sending.

Error Monitoring

Opt in on the tab you are debugging to capture console errors, exceptions, and network failures with focused stack context.

Browser Automation

Let your AI take control. Click buttons, fill forms, hover elements, and scroll pages through natural conversation.

Accessibility Auditing

Inspect ARIA roles, labels, and states instantly. Run comprehensive accessibility audits and get actionable reports.

Selected-Tab Events

Start and stop monitoring explicitly. Events are scoped to the tab and local terminal session you paired.

Works offline

Capture screenshots and select elements without a connection. Sessions stay active for 7 days and sync when you are back.

Page context

URL, title, and metadata travel with every interaction so the agent always knows which page you mean.

Works without a network

Sessions stay valid for 7 days offline. Capture and inspect locally, then sync when you reconnect.

  • Flights

    Keep using the panel at altitude.

  • Trains

    Work through tunnels without dropping the session.

  • Unstable Wi-Fi

    Progress is stored locally until sync.

  1. 01

    Sign in online

    Sign in once to receive a signed offline token.

  2. 02

    Work offline

    Use all features for up to 7 days without internet.

  3. 03

    Auto-sync

    Reconnect and the token refreshes automatically.

Signed tokens are verified locally. The session stays protected even without a network.

Three steps to connect

Install the extension, run setup, pair with your agent. Under two minutes.

01

Install

Add the Chrome extension from the Web Store.

# Install from Chrome Web Store
# or download from the official website
02

Setup

Run the signed npm setup wizard for your supported MCP client.

npx -y @continuallabs/agent-logger@1.1.0 setup
03

Connect

Open the side panel, enter the pairing code, and start sending context to your AI.

npx -y @continuallabs/agent-logger@1.1.0 doctor

Open the side panel and start sending context.

Works with your terminal

Direct injection for tmux and iTerm2. Clipboard fallback for Terminal.app and VS Code.

tmux

staged text

Context is staged without automatically executing a command

iTerm2

staged text

Context is staged without automatically pressing Enter

Terminal.app

Clipboard

Copies to clipboard - press Cmd+V to paste

VS Code

Clipboard

Copies to clipboard - paste into integrated terminal

Detection uses $TMUX, $ITERM_SESSION_ID, and $TERM_PROGRAM. The side panel shows Send or Copy based on injection support.

Lifetime access for $25

One payment. No subscription. Paid features and product updates while Agent Logger is supported.

  • Lifetime access
  • Supported product updates
  • Full feature access
  • Offline for 7 days
  • Up to 5 devices
  • Non-renewing purchase

Card required. Eligible first-time accounts are not charged at setup and receive 48 hours to cancel. An account that already used its trial is charged $25.00 USD once after card setup. No recurring fees. After payment, there is no discretionary change-of-mind refund; statutory rights remain.

$9 / month

$25

one-time payment

First eligible trial: one $25.00 USD charge at the displayed time, 48 hours after setup, unless cancelled. Prior trial: one immediate $25.00 USD charge after setup. Completed sales are final except where applicable law requires a remedy.

Not ready to buy

Eligible first-time accounts can try the paid features for 48 hours.

Card required. Cancel before the displayed charge time to pay nothing. Accounts that already used a trial do not receive another one.

Connect the browser to your agent

Install the extension and start sending context to Claude Code in seconds.

Chrome, Chromium, Brave, or Edge. Node.js 22.14+. macOS, Linux, or Windows.