MCP Server · Open ProtocolAI Browser Operations
AI Browser Operations
for Dolphin Anty
Connect MCP-compatible AI agents to browser profiles you own. Automate authorized QA, internal operations, and repeatable web workflows with visual interaction and recovery.
Practical Examples
See how it works in practice
Authorized QA and operations workflows running across browser profiles with Ornold MCP.
Prepare authorized browser workspaces
An AI agent connects to profiles your team owns, checks session readiness, and opens the approved workspace for a repeatable operation.
Prepare authorized browser workspaces
An AI agent connects to profiles your team owns, checks session readiness, and opens the approved workspace for a repeatable operation.Vision Mode
Interact visually when layouts change
Instead of relying on CSS selectors or XPath that break on any layout change, our Vision approach takes a screenshot and uses AI to detect every interactive element with pixel-precise coordinates.
1
ScreenshotCapture viewport from each browser in parallel
2
CLIP GroupingGroup identical pages to analyze each layout only once
3
Element DetectionDetect elements with box coordinates [x1, y1, x2, y2]
4
Coordinate ClickClick the center of the detected element using normalized coordinates
Открытый протокол
Что такое MCP?
Model Context Protocol (MCP) — открытый стандарт от Anthropic, который позволяет AI-агентам подключаться к внешним инструментам через единый универсальный интерфейс.
Один сервер — любой агент
Интеграция пишется один раз. Claude Code, Codex, Cursor, Windsurf и любой другой MCP-совместимый агент могут использовать её без изменений.Типизированные контракты
Каждый инструмент имеет типизированную схему — имя, описание, параметры. AI-агент обнаруживает инструменты автоматически и точно знает, как их вызвать.Почему мы выбрали MCP
Вместо создания отдельных плагинов для каждого AI-агента, MCP позволяет Ornold предоставить 30+ инструментов автоматизации браузера один раз — и они работают везде.Экосистема MCP
Комбинируйте с любыми MCP серверами
Ornold работает параллельно с другими MCP серверами в одном агенте. Автоматизация браузеров + мессенджеры + таблицы + базы данных — всё в одном разговоре.
Claude Code
Codex
Claude Desktop
Kilo Code
Roo Code
Windsurf
Cursor
Augment Code
Gemini CLI
MCP PROTOCOL
🌐
Ornold Browser MCPConnect through supported APIs or CDP and operate only profiles your organization owns or is authorized to manage.
➕Any CDP
🔌
Экосистема MCPПодключайте несколько MCP серверов к одному агенту. Каждый отвечает за свою область — Ornold управляет браузерами, другие работают со Slack, Excel, базами данных и прочим. Агент оркестрирует всё сам.
🔌 Any MCP
🌐 Ornold + ... =
Hover over a service to see how it works with OrnoldCore Features
Built for reliable browser operations
Tools for authorized QA and internal workflows, from vision-based interaction to controlled parallel execution across browser profiles.
Vision-First Approach
AI sees the page through screenshots, detects UI elements via vision analysis, and clicks by normalized coordinates. No fragile selectors.Controlled Parallel Execution
Run approved QA and operations steps across multiple connected profiles. Smart concurrency adapts to system resources.Approval Checkpoints
Pause sensitive steps for human review, preserve workflow state, and resume after an authorized operator confirms the action.Configurable Interaction
Use visual coordinates or structured page references with predictable timing, clear logs, and reproducible behavior.Auto-Sync & Recovery
Continuous status tracking. Automatic recovery for disconnected or behind browsers. No manual babysitting.Dual Interaction Mode
Vision mode (coordinates) as primary, accessibility snapshots (ref-based) as fallback. Two layers of reliability.Approval CheckpointsKeep sensitive steps
Keep sensitive steps
under operator control.
Pause a workflow for review, preserve the current browser state, and continue only after an authorized operator confirms the next action.
browser_parallel_wait_for
→ Waiting for approved page state
→ Workflow paused for operator review
→ Approval recorded in the audit trail
→ Approved sessions resumed
→ Final state verified
Manual approval
Pause and resume
Audit trail
Session status
Step confirmation
Structured logs
Targeted retry
Error recovery
MCP Tools
30+ tools for complete browser control
Every tool runs in parallel across all connected browsers. Target specific profiles with browserIds or tags.
browser_listList all connected browsers with status, URL, and tagsbrowser_statusSync status of all browsers — detect BEHIND or FAILED statesbrowser_parallel_navigateNavigate all browsers to the same URL in parallelbrowser_parallel_vision_analyze_groupedScreenshot + CLIP grouping + vision element detectionbrowser_parallel_click_normalized_boxClick element by normalized box coordinates from vision analysisbrowser_parallel_snapshotAccessibility tree snapshots with element refs for interactionbrowser_parallel_fillFill inputs with same or per-browser valuesbrowser_parallel_typeType text with configurable timing across approved browser sessionsbrowser_parallel_run_codeExecute Playwright code with per-browser template variablesbrowser_parallel_wait_forWait for text, element, or timeout across all browsersbrowser_run_recorded_flowReplay saved macros by ID across browser profilesStart automating in minutes
Add the MCP server to your agent config and connect browser profiles your team is authorized to operate.
1
Create an account & get your tokenSign up at mcp.ornold.com, go to API Tokens tab, and create a new token.
Create Free Account2
Install Node.js (if not installed)macOS (Homebrew)
brew install nodeWindows / Linux
nodejs.org → LTS3
Paste this prompt into your AI agentOpen your agent (Claude Code, Cursor, Codex, Windsurf, Cline, Claude Desktop) and paste the message below. The agent reads the public install guide on GitHub, asks which supported browser and interaction mode you use, then installs the integration.
Install Ornold MCP — AI browser automation for authorized workflows.
Full install guide for AI agents: https://github.com/ornold-ai/ornold-mcp/blob/main/AGENTS.md
Before installing, ask me:
1. Do I have an Ornold token (orn_sk_…)? If not, send me to https://mcp.ornold.com → Tokens → Create token.
2. Which supported browser do I use? (Linken Sphere / Dolphin Anty / Vision Browser / Wade X / another CDP browser) — this decides the connection flags.
3. Mode: "vision" (for dynamic interfaces) or "dom" (for structured pages). If I want both, set up two MCP servers so the tool list stays focused.
4. Is Node.js 18+ installed? Run "node -v"; if missing, point me to https://nodejs.org.
Then install Ornold MCP for the agent you're running in, using the install method and flags from AGENTS.md. Install the skill (npx mcp-ornold-browser --install-skill) where supported. Verify with browser_status. Use it only with systems, accounts, and data I own or am authorized to manage.4
Start automatingOnce the agent confirms
browser_status works, ask it to navigate, inspect pages, fill authorized forms, or record reusable QA and operations flows across your browser profiles.