Skip to main content
Glama

BrowserTrack 🔍

Local browser diagnostics shared with coding agents through MCP.

BrowserTrack is a lightweight, framework-agnostic local development daemon and Model Context Protocol (MCP) bridge. It automatically collects browser-side runtime errors, unhandled rejections, console logs, network failures, user breadcrumbs, and DOM snapshots—making them instantly accessible to AI coding assistants (Antigravity, Cursor, Claude Code) for rapid debugging and closed-loop verification.


⚡ Quick Start

1. Start the Local Daemon

npm install -g browsertrack
browsertrack start

This starts the local diagnostics server at http://127.0.0.1:7331 (WebSocket at ws://127.0.0.1:7331).

⚡ Zero-Config Auto-Start: If you configure BrowserTrack via MCP in your AI editor (Cursor, Antigravity, Claude Code), you do not need to manually run browsertrack start! The MCP server automatically launches and manages a single background daemon (singleton) on port 7331. Multiple windows or sessions share this single daemon without duplicate processes or port conflicts.

2. Connect Your Web Project

Option A: Zero-install script tag (Plain HTML / Vite / Next.js)

<script src="http://127.0.0.1:7331/client.js"></script>

Option B: NPM package (React / Vue / Svelte / Astro)

npm install -D browsertrack

In your main application entry point (e.g. main.ts or index.tsx):

import 'browsertrack/client';

Related MCP server: FlowLens MCP Server

🤖 Coding Agent (MCP) Setup

Add BrowserTrack to your MCP client configuration:

Cursor (~/.cursor/mcp.json) & Claude Code

{
  "mcpServers": {
    "browsertrack": {
      "command": "npx",
      "args": ["browsertrack", "mcp"]
    }
  }
}

Antigravity (~/.gemini/antigravity-ide/mcp_config.json)

{
  "mcpServers": {
    "browsertrack": {
      "command": "browsertrack",
      "args": ["mcp"]
    }
  }
}

🛠️ MCP Tools Exposed to AI Agents

📋 Incidents & Diagnostics

MCP Tool

Description

list_incidents

Lists grouped error incidents with occurrences, severity, and status.

get_incident

Retrieves concise, high-signal context (stack trace, breadcrumbs, network failures, last interacted element, error screenshot).

get_console

Retrieves recent console errors/warnings.

get_network_failures

Retrieves recent 4xx/5xx HTTP failures and aborted requests.

get_breadcrumbs

Returns user interaction timeline leading up to an error.

get_page_state

Queries live DOM/URL state from the active browser tab.

capture_element

Captures an on-demand screenshot of a specific element or full page.

📝 Visual Notes, Scenarios & Layout Inspector

MCP Tool

Description

list_notes

Lists visual annotations left on elements/regions (Alt+Click notes).

get_note

Retrieves complete note context (component source file & line, DOM selector, bounding rect, screenshot).

capture_note_context

Inspects live DOM bounding box, styles, and overflow on an element.

verify_note

Runs automated layout verification and geometry diffing with screenshot comparison.

get_note_verification

Retrieves latest layout verification verdict and screenshots.

resolve_note

Marks a visual note as RESOLVED.

reopen_note

Reopens a previously resolved visual note.

list_scenarios

Lists multi-step user reproduction flows.

get_scenario

Retrieves complete step-by-step reproduction flow.

🎛️ Session, Verification & Agent Guidelines

MCP Tool

Description

list_projects

Lists registered local development projects and mapped filesystem paths.

list_sessions

Lists active browser tabs connected over WebSocket.

verify_incident

Runs closed-loop verification: reloads browser, checks if error reoccurs, executes probes, captures after-screenshot, and returns verdict (VERIFIED, FAILED, INCONCLUSIVE).

get_verification

Returns latest verification result and before/after screenshot artifacts.

get_agent_guidelines

Retrieves full AI agent guidelines, workflows, and tool references.


💻 CLI Commands

Powered by mcponce:

browsertrack start             # Start local daemon
browsertrack start --verbose   # Verbose logging mode
browsertrack stop              # Stop background daemon
browsertrack status            # Check status & connected sessions

browsertrack projects          # List tracked projects
browsertrack project add myapp --origin http://localhost:5173 --path ~/projects/myapp

browsertrack errors            # List open runtime errors
browsertrack errors --project myapp

browsertrack clear             # Clear stored logs & incidents

# MCP & mcponce Developer Suite
browsertrack mcp               # Run MCP server over stdio (auto-boots singleton daemon)
browsertrack inspect           # Launch interactive zero-dependency Web Inspector & Playground
browsertrack mcp tools         # Inspect all 21 tool signatures and input schemas
browsertrack mcp call <tool>   # Test any MCP tool directly in terminal
browsertrack mcp analytics     # Real-time tool execution metrics and latency
browsertrack mcp install       # Auto-install MCP server to Claude Desktop or Cursor

🛡️ Security & Privacy

  • Localhost Only: Daemon strictly binds to 127.0.0.1.

  • Sensitive Key Redaction: Powered by @visulima/redact — Authorization, Cookie, Set-Cookie, passwords, API tokens, bearer tokens, JWTs, AWS credentials, credit cards, SSNs, and secret query parameters are automatically sanitized.

  • No Form Values: Form inputs (<input>, <textarea>) are never inspected or collected.

  • Zero Request/Response Body: Payload bodies are disabled by default.


🏗️ Architecture

Browser Client (Vanilla / React / Vue / Vite)
      ↕ WebSocket (ws://127.0.0.1:7331)
Local Daemon (SQLite Persistence + Verification Engine)
      ↕ MCP (stdio)
Coding Agents (Antigravity / Cursor / Claude Code)

🧪 Testing

Run the automated test suite:

npm test

Start the interactive fixture app for manual and end-to-end verification:

npx serve examples/test-app

Related MCP Connectors

Related MCP Servers

  • F
    license
    Not graded
    quality
    A
    maintenance
    Enables coding agents to access recorded browser flows—including user actions, network activity, console logs, storage, DOM events, and screen recordings—for debugging and regression testing. It lets agents inspect and analyze these flows to report bugs and generate tests without manually reproducing issues.
    -
  • F
    license
    Not graded
    quality
    A
    maintenance
    Enables coding agents to access full browser context from recorded user flows, including actions, network, console, storage, DOM events, and screen recordings. This supports in-depth debugging, bug reporting, and regression testing without manually reproducing issues or copying logs.
    -
  • F
    license
    Not graded
    quality
    A
    maintenance
    Enables coding agents to access recorded browser flows—including user actions, network activity, console logs, storage, DOM events, and screen recordings—for debugging, bug reporting, and regression testing without reproducing issues.
    -