MCP Browser Automation

A Model Context Protocol (MCP) host server that integrates multiple MCP servers with LLM capabilities. This skill provides browser automation via Playwright MCP with intelligent natural language command processing and dynamic tool selection.

What This Skill Does

This skill helps you set up, develop, and work with an MCP-based browser automation system that:

Architecture Overview

Core Components

Key Design Patterns

Instructions for AI Agent

When working with this MCP browser automation project, follow these steps:

1. Initial Setup and Verification

First, understand the project structure:

```bash

Read the project overview and architecture

Key files to examine:

- mcp_client.py (MCP client implementation)

- user_interface.py (CLI entry point)

- llm_config.py (LLM configuration)

- playwright_mcp_config.py (browser automation)

- setup.py (installation script)

- test_setup.py (test suite)

```

Verify the environment setup:

```bash

Check Python version (3.10+ required)

python --version

Verify virtual environment exists

ls venv/

Check .env file exists and has OPENAI_API_KEY

cat .env

```

If setup is needed:

```bash

Run initial setup (installs dependencies, Playwright, etc.)

venv/Scripts/python.exe setup.py

If .env doesn't exist, copy template

cp .env.example .env

Remind user to add OPENAI_API_KEY to .env

Verify installation

venv/Scripts/python.exe test_setup.py

```

2. Running the Application

For interactive mode with natural language processing:

```bash

venv/Scripts/python.exe user_interface.py --interactive

```

For single command execution:

```bash

Simple navigation

venv/Scripts/python.exe user_interface.py --command "open google.com"

Multi-step operations

venv/Scripts/python.exe user_interface.py --command "go to youtube and search for python"

```

For debugging:

```bash

LOG_LEVEL=DEBUG venv/Scripts/python.exe user_interface.py --interactive

```

3. Testing and Validation

Run the comprehensive test suite:

```bash

venv/Scripts/python.exe test_setup.py

```

Test individual components:

```bash

Test LLM configuration

venv/Scripts/python.exe -c "from llm_config import llm; print('LLM OK')"

Test MCP client

venv/Scripts/python.exe -c "from mcp_client import MCPClient; print('MCP Client OK')"

Test URL correction functionality

venv/Scripts/python.exe -c "from mcp_client import MCPClient; c = MCPClient({}); print(c.correct_url('goole.com'))"

```

4. Development Guidelines

**Code Patterns:**

**When modifying code:**

5. Understanding Natural Language Processing

The system processes natural language commands dynamically:

**Simple navigation with auto-correction:**

```

"open goole.com" # Auto-corrects to google.com

"visit youtub.com" # Auto-corrects to youtube.com

```

**Multi-step operations:**

```

"go to google and search for python tutorials"

"navigate to amazon and find laptops"

"open github and look for AI repositories"

```

**Complex browsing tasks:**

```

"visit news website and summarize the headlines"

"go to wikipedia and research machine learning"

```

6. Troubleshooting Common Issues

**If browser doesn't open:**

**If LLM calls fail:**

**If MCP tools aren't working:**

**If URL correction fails:**

7. Key Dependencies

**System Requirements:**

**Main Python Dependencies:**

8. Entry Points

Usage Examples

Setting Up a New Environment

```bash

Clone or navigate to repository

cd custom_mcp

Run setup

venv/Scripts/python.exe setup.py

Configure environment

cp .env.example .env

Add OPENAI_API_KEY=your-key-here to .env

Verify installation

venv/Scripts/python.exe test_setup.py

Start interactive mode

venv/Scripts/python.exe user_interface.py --interactive

```

Adding New Browser Automation Features

1. Review existing tools in `custom_mcp_server.py`

2. Add new tool definition following FastMCP patterns

3. Update `mcp_client.py` if new dynamic processing needed

4. Add test cases to `test_setup.py`

5. Update documentation in this file

Debugging Complex Workflows

```bash

Enable debug logging

LOG_LEVEL=DEBUG venv/Scripts/python.exe user_interface.py --interactive

Test specific command

LOG_LEVEL=DEBUG venv/Scripts/python.exe user_interface.py --command "your complex command here"

Check individual component

venv/Scripts/python.exe -c "from component import function; print(function())"

```

Important Constraints

Migration Notes

This implementation uses natural language processing instead of hard-coded commands:

| Old Approach | New Natural Language Approach |

|--------------|-------------------------------|

| `navigate https://google.com` | `"open google.com"` |

| `call-tool playwright browser_navigate` | `"go to google.com"` |

| Multiple separate commands | `"go to X and do Y"` (single command) |

The system dynamically interprets intent, corrects URLs, and chains operations automatically.