Donely Browser Plugin

Overview

Browser automation is one of the most powerful capabilities you can give an AI agent. The Donely Browser Plugin connects your OpenClaw instance to a cloud-hosted Chromium browser through the Chrome DevTools Protocol (CDP), enabling your agent to navigate the web, fill forms, scrape data, capture screenshots, and interact with any website — all without running a browser locally inside the container.

Unlike traditional browser automation frameworks such as Puppeteer or Selenium that require scripted selectors and coded workflows, OpenClaw browser automation is semantic. Your agent reads pages the way a human does — understanding context, recognizing buttons by meaning rather than CSS class names, and adapting when layouts change. Combined with Donely's cloud browser infrastructure, this creates an AI agent web scraping setup that is both powerful and resilient.

The plugin works through a relay architecture: your OpenClaw agent connects to the Donely Browser Service, which provisions an isolated Hyperbrowser cloud browser session with a persistent profile. The agent then communicates directly with the browser via CDP WebSocket, giving it full control over navigation, interaction, and data extraction.

How It Works

The Donely Browser Plugin operates through a multi-layer architecture designed for security, performance, and persistence:

OpenClaw Agent → GET /cdp/json/version → Donely Browser Service (browsers.donely.cloud)
                                              ├─ Verify gateway token (D1 database)
                                              ├─ Budget check (1 cent per session)
                                              ├─ Lookup or create Hyperbrowser profile
                                              └─ Create browser session → returns { wsEndpoint }
OpenClaw Agent ←── { webSocketDebuggerUrl } ←──┘
OpenClaw Agent ═══ Direct CDP WebSocket ═══► Hyperbrowser Cloud Browser

• Discovery Request — When the agent invokes the browser tool, OpenClaw sends a CDP discovery request to browsers.donely.cloud with the tenant's gateway token.
• Authentication & Budget — The Donely Browser Service verifies the token against the D1 database and checks that the tenant's API key has sufficient budget (each session costs 1 cent).
• Profile Resolution — The service looks up or creates a persistent Hyperbrowser profile for the tenant. This profile retains cookies, localStorage, sessionStorage, and network cache across sessions.
• Session Creation — A new Hyperbrowser cloud browser session is provisioned with the tenant's profile, including optional residential proxy routing based on the tenant's country.
• CDP Handoff — The service returns a webSocketDebuggerUrl pointing directly to Hyperbrowser. From this point, all browser communication flows directly between the OpenClaw container and the cloud browser — the relay service is no longer in the data path.
• Agent Control — OpenClaw's built-in browser tool takes over, using Playwright over CDP to navigate, click, type, screenshot, and extract data.

This architecture means the Donely Browser Service handles only authentication and session setup. The actual browser automation traffic flows directly to the cloud browser, minimizing latency and avoiding bandwidth bottlenecks.

Features

Persistent Browser Profiles

Live View Streaming

Session Recordings

File Downloads

File Uploads

Country-Based Proxy Routing

Budget-Controlled Sessions

Step-by-Step Setup

The Donely Browser Plugin is enabled through the platform's config patch system. For most tenants, it is enabled automatically during provisioning. To enable it manually:

1. Admin Panel Method

• Navigate to the Admin Panel at donely.ai/admin
• Find the tenant in the tenant list
• Open the Config Patches section
• Apply the "Enable Cloud Browser Plugin" patch
• The patch will:

2. Configuration Patch Details

The patch modifies the OpenClaw configuration (openclaw.json) with these settings:

{
  "plugins": {
    "entries": {
      "donely-browser": {
        "enabled": true,
        "config": {
          "browserServiceUrl": "https://browsers.donely.cloud",
          "gatewayToken": "<TENANT_GATEWAY_TOKEN>"
        }
      }
    }
  },
  "browser": {
    "enabled": true,
    "defaultProfile": "remote",
    "profiles": {
      "remote": {
        "cdpUrl": "https://browsers.donely.cloud/cdp/json/version?token=<TOKEN>"
      }
    }
  },
  "tools": {
    "alsoAllow": [
      "get_browser_live_view",
      "get_browser_recordings",
      "get_browser_downloads",
      "upload_to_browser",
      "list_browser_sessions"
    ]
  }
}

3. Verify the Setup

After applying the patch, verify the plugin is working:

• Open the tenant's chat interface on the Donely dashboard
• Ask the agent: "Open a browser and navigate to example.com"
• The agent should use the browser tool and describe what it sees
• Ask for a live view link to confirm the cloud browser is active

Agent Prompt

To get the best results from browser automation, include browser-specific instructions in your agent's BOOTSTRAP.md:

# Browser Automation Guidelines
You have access to a cloud browser through the Donely Browser Plugin. Use it for web tasks.

Before Browsing
Always call get_browser_live_view first and share the live view link with the user
Explain what you plan to do before navigating

During Browsing
Take screenshots after important actions to verify results
Use AI snapshots to understand page structure before interacting
Wait for pages to fully load (networkidle) before extracting data
Handle cookie consent banners and popups proactively

After Browsing
Call get_browser_recordings to share session recordings
If you downloaded files, use get_browser_downloads to share the download link
Summarize what you found or accomplished

Data Extraction
For tables, extract data into structured formats (CSV, JSON)
For multi-page data, navigate through pagination systematically
Verify extracted data against what's visible on the page

Security
Never enter credentials unless the user explicitly provides them and asks you to
Prefer the user logging in manually through the live view
Be cautious with financial transactions — always confirm with the user first

Available Tools

The Donely Browser Plugin registers five tools that complement OpenClaw's built-in browser tool:

get_browser_live_view

get_browser_recordings

• sessionId (optional) — Specific session ID. Defaults to the most recent session.

get_browser_downloads

• sessionId (optional) — Specific session ID. Defaults to the most recent session.

upload_to_browser

• filePath (required) — Absolute path to the file inside the container (e.g., /data/output/resume.pdf).

list_browser_sessions

• limit (optional) — Maximum number of sessions to return. Default: 10.

Use Cases

Web Scraping and Data Extraction

Form Filling and Submissions

Website Testing and QA

Screenshot and Visual Capture

Price and Content Monitoring

Social Media and Content Management

Configuration

OpenClaw Configuration (openclaw.json)

The browser plugin requires these configuration sections:

{
  "browser": {
    "enabled": true,
    "defaultProfile": "remote",
    "profiles": {
      "remote": {
        "cdpUrl": "https://browsers.donely.cloud/cdp/json/version?token=<GATEWAY_TOKEN>"
      }
    },
    "ssrfPolicy": "dangerouslyAllowPrivateNetwork"
  },
  "plugins": {
    "entries": {
      "donely-browser": {
        "enabled": true,
        "config": {
          "browserServiceUrl": "https://browsers.donely.cloud",
          "gatewayToken": "<GATEWAY_TOKEN>"
        }
      }
    }
  }
}

Environment Variables

The following are configured on the Donely Browser Service (not on individual tenants):

Plugin Manifest (openclaw.plugin.json)

{
  "id": "donely-browser",
  "name": "Donely Cloud Browser",
  "description": "Cloud browser tools — live view, recordings, downloads, uploads via Hyperbrowser.",
  "configSchema": {
    "type": "object",
    "properties": {
      "browserServiceUrl": { "type": "string" },
      "gatewayToken": { "type": "string" }
    }
  }
}

Real-World Examples

Example 1: Competitive Price Monitoring

User: Check the prices for "wireless earbuds" on Amazon and Best Buy, 
      then create a comparison spreadsheet.
Agent: I'll browse both sites and extract pricing data. Here's your live 
       view link so you can watch: [live view URL]
       [Navigates to Amazon, searches "wireless earbuds", extracts top 10 
       results with names, prices, ratings]
       [Navigates to Best Buy, repeats the process]
       [Creates CSV file with columns: Product, Amazon Price, Best Buy Price, 
       Difference]
Agent: Done! I found 10 matching products. Here's the comparison:
       - AirPods Pro: $199 (Amazon) vs $199 (Best Buy)
       - Sony WF-1000XM5: $248 (Amazon) vs $279 (Best Buy) — $31 cheaper on Amazon
       ...
       The CSV is saved at /data/output/price-comparison.csv
       Session recording: [recording URL]

Example 2: Automated Job Application

User: I uploaded my resume to /data/uploads/resume.pdf. Apply to the 
      3 most recent "Software Engineer" positions on LinkedIn.
Agent: I'll open LinkedIn and search for positions. Here's the live view: 
       [live view URL]
       Note: I'll need you to log in via the live view first, since I 
       shouldn't handle your credentials directly.
User: Done, I'm logged in.
Agent: [Searches "Software Engineer", filters by "Past 24 hours"]
       [For each of 3 listings: clicks Apply, uploads resume via 
       upload_to_browser, fills application fields, submits]
Agent: Applied to 3 positions:
       1. Senior Software Engineer at Acme Corp — Applied
       2. Software Engineer II at TechStart — Applied  
       3. Full Stack Engineer at DataFlow — Applied
       
       Session recording: [recording URL]
       Downloaded confirmation pages: [download URL]

Example 3: Website QA Testing

User: Test the checkout flow on staging.mystore.com. Add an item to cart, 
      go through checkout with test payment, and report any issues.
Agent: Starting QA test. Live view: [live view URL]
       [Navigates to staging.mystore.com]
       [Browses product catalog, adds "Test Widget" to cart]
       [Proceeds to checkout, fills shipping form with test data]
       [Enters test card number 4242424242424242]
       [Submits order]
       Issue found: The "Place Order" button shows a loading spinner 
       but doesn't redirect to the confirmation page. After 30 seconds, 
       it shows "Network Error". The order was NOT completed.
       Screenshots captured at each step.
       Session recording: [recording URL]
       
       Summary: Checkout flow fails at payment submission. Likely a 
       backend API timeout or misconfigured payment gateway on staging.

Security Considerations

Session Isolation

Token-Based Authentication

Budget Controls

No Credential Storage

Sandboxed Execution

Approval Gates

Best Practices and Troubleshooting

Best Practices

• Always share the live view link before starting browser tasks. Transparency builds trust and lets the user intervene if needed.
• Use persistent profiles wisely — log into services manually through the live view rather than giving the agent your credentials.
• Set reasonable budgets on API keys to control browser session costs. At 1 cent per session, 100 sessions cost only $1.
• Include browser instructions in BOOTSTRAP.md to guide the agent's browsing behavior consistently.
• Use recordings for audit trails — every session is recorded, providing full transparency for compliance or debugging.
• Prefer AI snapshots over screenshots for data extraction — they are faster and more structured.
• Wait for networkidle before extracting data from dynamic pages to ensure all content has loaded.

Troubleshooting