Multi-Profile Chrome Automation for AI Agents
How WebBridge orchestrates multiple Google Chrome profiles simultaneously with isolated cookies, deterministic port routing, and two-dimensional parallelism.
127.0.0.1:10086 inspects the top-level "profile" argument of incoming MCP requests and transparently forwards them to the correct profile. Sessions, logins, and cookies remain 100% isolated.
1. The Multi-Profile Fleet Architecture
Standard browser automation scripts struggle with multiple profiles because Chrome locks profile directories against multiple simultaneous instances. WebBridge works with Chrome's native multi-profile engine rather than fighting it:
Key architectural properties:
- Deterministic Hashed Ports: Each profile directory name is hashed to a stable port in the range 10000–15000 (e.g.
Work→ 13222). There is never a port collision. - LevelDB Direct Configuration: The
awb connectcommand writes the daemon URL directly to the extension's local LevelDB storage, eliminating manual popup configuration clicks. - Session Tab Groups: Each agent session organizes its opened tabs into a visible Chrome tab group named after the session (e.g.
agent:audit), completely separated from your daily tabs.
2. CLI Management & Profile Discovery
Discovering and bringing up your fleet requires only two CLI commands:
# 1. Discover all Chrome profiles on your system
awb profiles
# Output:
# Found 3 profiles:
# - Profile: "Work" (Port: 13222, Daemon: UP)
# - Profile: "Personal" (Port: 13336, Daemon: UP)
# - Profile: "QA-Test" (Port: 14102, Daemon: UP)
# 2. Boot the entire fleet behind router 10086
awb up "Work" "Personal" "QA-Test"3. Two-Dimensional Fleet Concurrency (N × M)
WebBridge offers two distinct dimensions of parallelism:
- Dimension 1 (Cross-Profile): N different Chrome profiles running simultaneously, each with distinct accounts and IP sessions.
- Dimension 2 (Per-Tab CDP): M concurrent tabs per profile running independent CDP sessions attached via a Map.
This delivers total concurrency of N × M. In reproducible live benchmarks, 5 tabs in a single profile evaluate in ~3.5 seconds in parallel vs 9.3 seconds serially:
| Execution Model | 1 Tab | 3 Tabs | 5 Tabs | Profile Isolation |
|---|---|---|---|---|
| Legacy Browser Bridges | ~1.9s | ~5.8s (serial queue) | ~9.3s (serial queue) | 1 profile at a time |
| WebBridge Fleet (AWB) | ~1.9s | ~2.2s (parallel) | ~3.5s (parallel) | N isolated profiles |
4. Legitimate Use Cases & Safety Guardrails
Multi-profile automation is designed for authorized developer workflows:
- Role-Based Access Testing: Verifying permissions simultaneously as Admin in Profile 1 and Standard User in Profile 2.
- Work vs Personal Isolation: Allowing an agent to access company GitHub in Profile "Work" without exposing personal Google Drive in Profile "Personal".
- Client Account Separation: Maintaining strictly isolated browser states for independent client consulting engagements.
Guardrails: WebBridge is not designed for platform evasion, multi-account mass botting, or circumventing website terms of service. All profiles share your local machine's egress IP address and hardware fingerprint.
Frequently Asked Questions
{ profile: "Work" } and { profile: "Personal" }), an agent can read an email in one profile while updating a pull request or verifying a dashboard in another, without logging in and out.