Mymanus

MyManus Plugin for Claude Code

Transform Claude Code into an autonomous agent with MyManus capabilities - planning, reasoning, execution, and evaluation - all without writing a single line of code.

Overview

This plugin brings the powerful agentic behavior of MyManus to Claude Code, Anthropic's official CLI for Claude. It combines MyManus's structured agent loop methodology with Claude Code's robust development tools to create a truly autonomous AI assistant.

What You Get

From MyManus:

• 🎯 Structured agent loop (Plan → Execute → Observe → Iterate)
• 🧠 Multi-module architecture (Planner, Knowledge, Datasource)
• 🌐 Browser automation capabilities via Playwright
• 📋 Systematic task planning and progress tracking
• 📝 Professional writing and research workflows
• 🔄 Autonomous error handling and strategy adaptation

From Claude Code:

• ⚡ Fast, native development tools (Bash, Read, Write, Edit, Glob, Grep)
• 🔍 Powerful web research (WebFetch, WebSearch)
• 🤖 Specialized agents for complex tasks
• 📊 Built-in task management (TodoWrite)
• 🔌 MCP server integration
• 🖥️ Cross-platform support (Linux, macOS, Windows)

Together: A powerful, autonomous AI agent that can research, code, analyze data, write reports, and automate complex workflows - all while keeping you informed of progress every step of the way.

Key Features

1. Autonomous Agent Behavior

Unlike standard Claude interactions, the MyManus plugin enables true autonomous operation:

• Plans complex tasks independently
• Executes multi-step workflows without constant supervision
• Adapts strategies when encountering obstacles
• Validates results before marking tasks complete

2. Browser Automation

Full web automation via Playwright MCP server:

• Navigate complex websites
• Fill forms and interact with dynamic content
• Handle JavaScript-heavy applications
• Perform multi-step web research workflows
• Close cookie banners and popups automatically

3. Structured Task Management

Built-in planning and tracking:

• Automatic task breakdown for complex requests
• Real-time progress updates via TodoWrite
• Clear status reporting (pending → in_progress → completed)
• Transparent strategy changes and error handling

4. Professional Research & Writing

Specialized workflows for information work:

• Multi-source fact-checking and validation
• In-depth research with comprehensive documentation
• Long-form article and report writing
• Proper citation and reference management

5. Systematic Coding

Organized development practices:

• Project structure planning
• Code organization and documentation
• Testing and validation
• Error handling and debugging

Installation

Prerequisites

1. Claude Code: Install from claude.com/claude-code
2. Node.js: Required for Playwright MCP server (v18 or later recommended)
3. Git: For accessing the plugin repository

Quick Install (Recommended)

The easiest way to install the MyManus plugin is through Claude Code's plugin system:

1. Add the MyManus marketplace (one-time setup):

   /plugin marketplace add https://github.com/emsi/MyManus.git
   

2. Install the plugin:

   /plugin install mymanus@mymanus
   

3. Restart Claude Code to activate the plugin

4. Verify installation:

   /help
   

   The MyManus skill should now be available for Claude to use automatically.

Manual Install (Alternative)

If you prefer to install manually or for development purposes:

Step 1: Clone the Repository

git clone https://github.com/emsi/MyManus.git

Step 2: Add as Local Plugin

In Claude Code, add the local plugin:

/plugin marketplace add /path/to/MyManus
/plugin install mymanus@local

Step 3: Configure Playwright MCP Server

For Linux/macOS:

1. Ensure Node.js and npx are installed:

   node --version
   npx --version
   

2. Add Playwright MCP server to Claude Code's MCP configuration:

   • Open Claude Code settings/preferences
   • Navigate to MCP Servers section
   • Add the following configuration:

   {
     "playwright": {
       "runtime": "node",
       "command": "npx",
       "args": [
         "-y",
         "@automatalabs/mcp-server-playwright"
       ]
     }
   }
   

3. Linux/WSL2 Only: If you're using Linux or WSL2, you may need to set the DISPLAY variable:

   {
     "playwright": {
       "runtime": "node",
       "command": "npx",
       "args": ["-y", "@automatalabs/mcp-server-playwright"],
       "env": {
         "DISPLAY": ":0"
       }
     }
   }
   

   Ensure you have an X server running (on WSL2, use VcXsrv or X410).

For Windows:

Windows users typically don't need the DISPLAY environment variable. Use this configuration:

{
  "playwright": {
    "runtime": "node",
    "command": "npx",
    "args": [
      "-y",
      "@automatalabs/mcp-server-playwright"
    ]
  }
}

How the MyManus Skill Works

The MyManus plugin is implemented as a Claude Code Skill, which means:

• Auto-Activation: Claude automatically invokes the skill when it detects tasks requiring autonomous agent behavior, complex planning, research, or multi-step workflows
• Context-Aware: The skill description helps Claude determine when to use MyManus capabilities
• On-Demand Loading: The skill only loads when needed, keeping your conversations efficient
• Seamless Integration: Works alongside Claude Code's existing features without conflicts

You don't need to manually activate the skill - Claude will use it when appropriate based on your request.

Verify Installation

Test the installation with examples that trigger the MyManus skill:

Test 1: Simple browser automation

Use the browser to navigate to example.com and tell me what you see.

Test 2: Research task (triggers autonomous behavior)

Research the latest developments in Rust web frameworks and create a comparison table.

Claude should:

• Automatically create a TodoWrite plan
• Perform systematic research
• Provide comprehensive results

Usage

Basic Usage

Once installed, simply use Claude Code normally. Claude will automatically invoke the MyManus skill for tasks that benefit from autonomous agent behavior.

Example simple task:

Research the latest Python web frameworks and create a comparison table.

Claude will:

1. Create a todo list with research steps
2. Use web search to find current information
3. Visit multiple sources for comprehensive data
4. Compile findings into a structured table
5. Mark each step complete as it progresses

Example complex task:

Build a simple web app that fetches weather data and displays it with charts.

Claude will:

1. Plan the architecture (backend, frontend, API integration)
2. Create project structure
3. Write backend code for API calls
4. Create frontend with visualization
5. Test the application
6. Provide deployment instructions
7. Keep you updated throughout via TodoWrite

Advanced Features

1. Web Research with Browser Automation

For complex web interactions:

Use the browser to find the top 5 machine learning courses on Coursera,
including their ratings, duration, and enrollment numbers.

Claude will use Playwright to:

• Navigate to Coursera
• Handle cookie banners
• Search for ML courses
• Extract detailed information
• Compile results systematically

2. Long-Form Writing

For research reports and articles:

Write a 3000-word article about the history of artificial intelligence,
with proper citations and references.

Claude will:

• Research from multiple authoritative sources
• Create an outline
• Write individual sections
• Cite sources properly
• Compile a complete, well-structured article

3. Code Projects

For development tasks:

Create a Python script that monitors a directory for new files
and automatically backs them up to a specified location.

Claude will:

• Plan the implementation
• Create proper project structure
• Write well-documented code
• Add error handling
• Create tests
• Provide usage documentation

Tips for Best Results

1. Be Specific: Clear task descriptions lead to better planning
2. Allow Autonomy: Trust the agent loop to handle details
3. Check Progress: Watch TodoWrite updates to track progress
4. Provide Feedback: Correct course when needed, but let Claude adapt
5. Complex Tasks: Break huge projects into major phases yourself, let Claude handle each phase autonomously

Differences from Standalone MyManus

Feature	Standalone MyManus	MyManus Plugin for Claude Code
Platform	Claude Desktop with custom setup	Claude Code CLI
Installation	Manual sandbox setup required	Simple plugin installation
File Tools	MCP filesystem server	Native Read/Write/Edit tools
Shell Access	MCP shell server (sandboxed)	Native Bash tool
Browser	Playwright MCP	Playwright MCP + WebFetch
Task Tracking	Manual todo.md files	TodoWrite tool
Sandboxing	Custom claude_sandbox.sh	Claude Code's built-in security
Web Search	Browser-based only	Native WebSearch + browser
Best For	Desktop users, visual workflows	Developers, CLI enthusiasts

When to Use Each

Use Standalone MyManus if:

• You prefer desktop GUI interaction
• You need visual project selection
• You want the original sandboxed environment
• You're already using Claude Desktop

Use MyManus Plugin for Claude Code if:

• You prefer CLI/terminal interfaces
• You're a developer or power user
• You want native integration with dev tools
• You need faster tool execution
• You want to version control your projects with git

Troubleshooting

Playwright MCP Not Working

Symptom: Browser doesn't launch or Playwright tools are unavailable

Solutions:

1. Verify Node.js installation: node --version
2. Check MCP configuration is saved correctly
3. Restart Claude Code completely
4. Check logs for MCP server errors
5. On Linux/WSL2: Ensure X server is running and DISPLAY is set
6. Try manual installation: npx -y @automatalabs/mcp-server-playwright

System Prompt Not Loading

Symptom: Claude doesn't exhibit autonomous behavior

Solutions:

1. Verify system-prompt.md is in the correct location
2. Check Claude Code project settings
3. Ensure the prompt isn't overridden by user settings
4. Restart Claude Code after adding the prompt

Tasks Not Being Tracked

Symptom: No TodoWrite updates appear

Solutions:

1. Verify TodoWrite tool is available in Claude Code
2. Check if the task is complex enough (simple tasks don't need tracking)
3. Ensure the system prompt is loaded correctly

Browser Shows DISPLAY Error (Linux/WSL2)

Symptom: "Error: DISPLAY environment variable not set"

Solutions:

1. Install X server (VcXsrv on Windows for WSL2, or native on Linux)
2. Set DISPLAY variable: export DISPLAY=:0
3. Add DISPLAY to MCP config env section
4. Test X server: xclock (should show a clock window)

Examples

See the examples/ directory for detailed use case demonstrations:

• Research: Multi-source fact-checking and report writing
• Coding: Building complete applications with testing
• Analysis: Data processing and visualization
• Automation: Workflow automation and scripting

Configuration Options

Customizing the System Prompt

You can modify system-prompt.md to adjust behavior:

• Language Settings: Change default working language
• Writing Style: Adjust verbosity and formatting preferences
• Tool Preferences: Prioritize certain tools over others
• Error Handling: Change how errors are reported and handled

Adding More MCP Servers

The plugin is designed to work with additional MCP servers. See mcp-config.json for optional servers and instructions on adding new ones.

Popular additions:

• Database MCP: For direct database access
• API MCP: For specific API integrations
• Cloud Services MCP: For AWS, GCP, Azure operations

Contributing

Found a bug or have a suggestion? Please contribute!

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Submit a pull request

License

This plugin inherits the AGPL-3.0 license from MyManus. See LICENSE for details.

Support

• Issues: Report bugs at GitHub Issues
• Discussions: Join discussions about features and use cases
• Documentation: Full MyManus documentation at main README

Acknowledgments

• MyManus: Original project by @emsi
• Claude Code: Official CLI by Anthropic
• MCP: Model Context Protocol by Anthropic
• Playwright: Browser automation by Microsoft

---

Ready to transform Claude Code into an autonomous agent? Start with the Installation section above!