ai-mcp-terminal

AI-MCP Terminal

🚀 Multi-threaded terminal management for AI assistants with real-time web monitoring

Solve terminal blocking issues - Commands run async, never block AI operations. Monitor up to 100 concurrent terminals with intelligent cleanup and system tracking.

---

✨ Key Features

Core Capabilities

• 🚀 Async Execution - Commands never block AI operations
• 🔢 Multi-Threading - 100 concurrent terminals with ThreadPoolExecutor
• 🧹 Auto Cleanup - Smart idle session detection & memory management
• ⚡ Batch Operations - Execute across multiple terminals simultaneously
• 📊 Web Monitor - Real-time xterm.js interface with system stats

Smart Execution (v1.0.52+)

• 🔗 Workflow Engine - Execute tasks with dependencies (DAG support)
• ⏳ Smart Waiting - Block until specific tasks complete
• 📝 Sequential Execution - Run commands in strict order
• 🔄 Auto Retry - Automatic retry on transient failures
• 📂 Project Lock - Terminals always start in project directory

Platform Support

• 🐧 WSL Priority - Auto-detect WSL bash on Windows (preferred)
• 🌐 UTF-8 Support - Proper encoding, no garbled text
• 🛑 Anti-Loop Protection - Prevents AI from getting stuck in query loops

---

🚀 Quick Start (1 Minute)

Step 1: Add MCP Configuration

Add to your Cursor/Cline MCP settings:

{
  "mcpServers": {
    "ai-mcp-terminal": {
      "command": "uvx",
      "args": ["ai-mcp-terminal"],
      "env": {}
    }
  }
}

Step 2: Restart IDE

Step 3: Start Using

In Cursor:

Create 3 terminals and run system checks in parallel

AI will use create_batch for true concurrency!

Browser auto-opens → http://localhost:8000 → View all terminals in real-time!

---

📊 Web Interface

Auto-opens at http://localhost:8000

Features:

• 📺 Real-time xterm.js terminals
• 📊 CPU/Memory/System stats
• 🔄 Live output streaming
• 🎯 Click to expand terminals
• 🛑 Shutdown server button

---

🛠️ Available MCP Tools

Batch Tools (Recommended)

Smart Execution Tools (v1.0.52+)

Single Tools (Use batch tools instead!)

Why batch tools?

• 10x faster (parallel execution)
• 1 call instead of 10 calls
• Non-blocking design

---

🎯 Use Cases

Multi-Service Development

User: "Start frontend, backend, and database"

AI calls:
create_batch(sessions=[
  {name: "frontend", cwd: "./web", initial_command: "npm run dev"},
  {name: "backend", cwd: "./api", initial_command: "python app.py"},
  {name: "db", cwd: "./", initial_command: "docker-compose up"}
])

Result: 3 services start simultaneously, web interface shows all

System Information Gathering

User: "Check system info"

AI calls:
create_batch(sessions=[
  {name: "cpu", cwd: ".", initial_command: "wmic cpu get name"},
  {name: "mem", cwd: ".", initial_command: "wmic memorychip get capacity"},
  {name: "disk", cwd: ".", initial_command: "wmic logicaldisk get size,freespace"},
  {name: "os", cwd: ".", initial_command: "systeminfo"}
])

Later:
get_batch_output(session_ids=["cpu", "mem", "disk", "os"])

Result: All info gathered in parallel, 4x faster than serial

PyPI Release Workflow (v1.0.52+)

User: "Release to PyPI"

AI calls:
execute_workflow(tasks=[
  {
    name: "clean",
    session_id: "build",
    command: "rm -rf dist build *.egg-info"
  },
  {
    name: "build",
    session_id: "build",
    command: "python -m build",
    depends_on: ["clean"]
  },
  {
    name: "upload",
    session_id: "upload",
    command: "twine upload dist/*",
    depends_on: ["build"],
    retry: true,
    max_retries: 3
  }
])

Result: 
- Clean executes first
- Build waits for clean to complete
- Upload waits for build, retries on failure
- Tasks run in parallel when possible

Smart Retry for Network Operations

User: "Download and install dependencies"

AI calls:
execute_with_retry(
  session_id: "npm_install",
  command: "npm install",
  max_retries: 3,
  retry_delay: 2.0
)

Result:
- Attempt 1 fails (network error)
- Wait 2 seconds
- Attempt 2 fails  
- Wait 2 seconds
- Attempt 3 succeeds ✓

---

⚙️ Configuration

Optional environment variables:

{
  "mcpServers": {
    "ai-mcp-terminal": {
      "command": "uvx",
      "args": ["ai-mcp-terminal"],
      "env": {
        "AI_MCP_PREFERRED_SHELL": "bash"
      }
    }
  }
}

Shell Priority:

• Windows: WSL bash (🐧) → Git Bash (🐚) → powershell → cmd
• macOS: zsh → bash → sh
• Linux: bash → zsh → sh

v1.0.52: WSL now displays with penguin icon (🐧) in web interface, Git Bash with shell icon (🐚)

---

🔧 Installation Options

Option 1: UVX (Recommended)

```json
{

"command": "uvx", "args": ["ai-mcp-terminal"] }

**No installation needed!** UV handles everything.

### Option 2: PIPX

```bash
pipx install ai-mcp-terminal

```json
{

"command": "ai-mcp-terminal" }

### Option 3: PIP

```bash
pip install ai-mcp-terminal

{
      "command": "python",
  "args": ["-m", "src.main"]
}

---

🛡️ Anti-Loop Protection

Problem: AI gets stuck querying terminal repeatedly

Solution: Built-in query counter

• Query 1-2: Normal
• Query 3-4: ⚠️ Warning + stop instruction
• Query ≥5: 🔪 Auto-terminate process

Result: AI never loops, always proceeds with tasks

---

🚦 How AI Should Use This

✅ Correct Pattern

Dialog 1:
User: "Deploy React app"
AI: 
  1. create_batch(...) 
  2. Reply: "Deploying in background..."
  3. END conversation

Dialog 2 (later):
User: "Is it done?"
AI:
  1. check_completion(...)
  2. Reply: "Still running..." or "Done!"
  3. END conversation

❌ Wrong Pattern (Fixed by protection)

Dialog 1:
User: "Deploy React app"
AI:
  1. execute_command(...)
  2. get_output(...) → running
  3. get_output(...) → running  [Query 2]
  4. get_output(...) → running  [Query 3 - WARNING]
  5. get_output(...) → running  [Query 4]
  6. get_output(...) → AUTO-KILLED [Query 5]
  7. Error: "Loop detected, process terminated"

---

📁 Project Structure

ai-mcp-terminal/
├── src/
│   ├── main.py              # Entry point
│   ├── mcp_server.py        # MCP protocol handler (30+ tools)
│   ├── terminal_manager.py  # Terminal management (3400+ lines)
│   ├── web_server.py        # FastAPI + WebSocket
│   ├── key_mapper.py        # Keyboard interaction support
│   └── static/              # Web UI (xterm.js)
├── docs/                    # Documentation (15+ guides)
├── examples/                # Usage examples
├── CHANGELOG.md             # Detailed version history
├── README.md
├── LICENSE
└── pyproject.toml

---

🔧 Troubleshooting

Web Interface Not Opening

Solution: Visit http://localhost:8000 manually

Port Already in Use

Solution:

1. Auto-finds next available port
2. Or click shutdown in existing interface

AI Keeps Using Single Tools

Solution:

1. Restart IDE (MCP caches tool definitions)
2. Check tool descriptions loaded correctly

---

📄 License

MIT License - see LICENSE

---

🤝 Contributing

Contributions welcome! See CONTRIBUTING.md

---

🔗 Links

• PyPI: https://pypi.org/project/ai-mcp-terminal/
• GitHub: https://github.com/kanniganfan/ai-mcp-terminal
• Issues: https://github.com/kanniganfan/ai-mcp-terminal/issues
• Changelog: CHANGELOG.md

---

🆕 What's New in v1.0.53

🎯 Production-Ready Improvements

Based on real PyPI release testing, v1.0.53 brings battle-tested improvements that solve actual production issues:

🔍 Enhanced Debugging

• Detailed Statistics: Every command returns output_bytes, output_lines, execution_time, encoding_used
• Clear Status: Explicit success: true/false instead of ambiguous exit_code: null
• No More Guessing: Know exactly what happened with every command

🛡️ Smart Error Prevention

• Shell Type Detection: Warns when PowerShell command sent to Bash terminal (and vice versa)
• Quick Fix Suggestions: Provides exact commands to fix common errors
• 7 Error Categories: PyPI duplicates, encoding errors, permissions, network, syntax, etc.

🌐 Zero-Config UTF-8 (Windows)

• Auto Setup: Sets PYTHONIOENCODING=utf-8 and PYTHONUTF8=1 automatically
• No More Encoding Errors: twine, pip, and other Python tools just work
• 80% Fewer Errors: Eliminates common UnicodeEncodeError issues

🔄 Intelligent Batch Execution

• Smart Queueing: Same terminal → sequential, different terminals → concurrent
• Zero Race Conditions: No more "upload before build finishes" issues
• Maximum Efficiency: Still fully concurrent across different terminals

Previous Features (v1.0.52)

• ✨ execute_workflow() - DAG-based task orchestration
• ⏳ wait_until_complete() - Smart blocking wait
• 📝 execute_sequence() - Sequential execution with error handling
• 🔄 execute_with_retry() - Automatic retry mechanism

See CHANGELOG.md for complete details.

---

Made with ❤️ for AI Assistants

If this helps you, please give it a ⭐ star!