mcptui

---

🚀 How to Run mcptui (Quickstart)

Choose whichever method you prefer:

⚡ Method 1: Zero-Install (Run directly via uvx)

If you have uv installed, launch mcptui instantly without cloning or setting up virtual environments:

# Test drive the built-in demo server:
uvx --from git+https://github.com/kimarsen/mcptui mcptui demo

# Connect directly to your own server script:
uvx --from git+https://github.com/kimarsen/mcptui mcptui run "python my_server.py"

📦 Method 2: Clone & Run in PowerShell / Bash (Standard Method)

# 1. Clone the repository
git clone https://github.com/kimarsen/mcptui.git
cd mcptui

# 2. Launch interactive demo in 1 command
uv run mcptui demo

(If you don't have uv yet, install it via powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex" or use standard pip install -e .)

🛠 Method 3: Install Globally as a CLI Tool

# With uv tool:
uv tool install --from git+https://github.com/kimarsen/mcptui mcptui

# Or with pipx:
pipx install git+https://github.com/kimarsen/mcptui.git

# Now 'mcptui' is available everywhere in your terminal:
mcptui demo
mcptui run "python my_server.py"
mcptui servers

📥 Method 4: GitHub Releases (.whl packages)

Pre-built wheels and source distributions are automatically packaged on every release. You can download them directly from the Releases page and install via pip install mcptui-*.whl.

---

🎯 Why mcptui?

Building and debugging MCP (Model Context Protocol) servers is rapidly becoming a core workflow for AI developers. But today, testing servers is painful:

• You have to edit claude_desktop_config.json and restart Claude Desktop repeatedly.
• You waste money and time on LLM token costs just to check if a simple function runs without syntax errors.
• Official web inspectors require Node.js, launch heavy browser tabs (localhost:5173), and don't fit into terminal workflows.

mcptui fixes this completely. It gives you a dedicated, keyboard-driven Terminal UI (TUI) playground to test tools, inspect schemas, browse resources, preview prompts, and watch live JSON-RPC 2.0 traffic in real time with zero token costs.

 ┌────────────────────────────────────────────────────────────────────────┐
 │ mcptui | Postman for MCP                [* mcptui-demo] (18ms)         │
 ├────────────────────────────────────────────────────────────────────────┤
 │ [1] Tools   [2] Resources   [3] Prompts   [4] Protocol Logs   [5] History │
 ├─────────────────────────┬──────────────────────────────────────────────┤
 │ Filter tools...         │ TOOL: calculate                              │
 │ ─────────────────────── │ Safe arithmetic calculation                  │
 │ > calculate             │                                              │
 │ > weather               │ Arguments (JSON)                             │
 │ > system_info           │ ┌──────────────────────────────────────────┐ │
 │ > text_analyzer         │ │ { "expression": "100 * 2 + 42" }         │ │
 │ > mock_query            │ └──────────────────────────────────────────┘ │
 │                         │ [Execute (Ctrl+Enter)] [Copy Result]  OK 18ms │
 │                         │                                              │
 │                         │ Response                                     │
 │                         │ ┌──────────────────────────────────────────┐ │
 │                         │ │ { "result": 242, "status": "success" }   │ │
 │                         │ └──────────────────────────────────────────┘ │
 └─────────────────────────┴──────────────────────────────────────────────┘

---

🛠 Building Your First MCP Server in 2 Minutes

Want to create your own MCP server and test it with mcptui? Here is the complete minimal guide:

Step 1: Install the MCP Python SDK

pip install mcp
# or with uv:
uv add mcp

Step 2: Write your server (my_server.py)

import asyncio
from mcp.server.mcpserver import MCPServer

# 1. Initialize server
server = MCPServer("my-first-server")

# 2. Expose a tool that AI agents can execute
@server.tool(name="calculate_discount", description="Calculate price after applying percentage discount")
def calculate_discount(price: float, discount_percent: float) -> str:
    discounted = price * (1.0 - (discount_percent / 100.0))
    return f"Original: ${price}, Discount: {discount_percent}%, Final Price: ${round(discounted, 2)}"

# 3. Expose a static data resource
@server.resource(uri="store://catalog", name="Store Catalog", mime_type="application/json")
def get_catalog() -> str:
    return '{"items": ["Laptop", "Mouse", "Keyboard"], "status": "in_stock"}'

# 4. Run over standard input/output (stdio)
if __name__ == "__main__":
    asyncio.run(server.run_stdio_async())

Step 3: Test and Debug Live in mcptui

uv run mcptui run "python my_server.py"

Your new tool calculate_discount and resource store://catalog will instantly appear in the TUI! Fill in parameters, press Ctrl+Enter, and verify the output in 20ms.

Step 4: Connect to Claude Desktop or Cursor

Save it into your local mcp.json or %APPDATA%\Claude\claude_desktop_config.json:

{
  "mcpServers": {
    "my-store-server": {
      "command": "python",
      "args": ["path/to/my_server.py"]
    }
  }
}

(Or simply run mcptui add my-store-server "python path/to/my_server.py" and mcptui will register it for you!)

---

✨ Features

• 🛠 Interactive Tool Workspace:

  • Auto-generates clean JSON parameter templates from tool inputSchema.
  • In-terminal JSON editor with error validation.
  • One-click execution (Ctrl+Enter) with live duration & latency measurement (OK 18ms).
  • Pretty-printed, syntax-highlighted output with clipboard copying.

• 📚 Resource Explorer:

  • Browse available server resources and dynamic URI templates.
  • Read resources with automatic syntax detection (Markdown, JSON, Plain Text).

• 💬 Prompt Inspector:

  • Test prompt templates with dynamic argument substitution.
  • Preview generated message sequences (System, User, Assistant) before sending them to LLMs.

• 🕵️ Live JSON-RPC 2.0 Traffic Inspector:

  • Intercepts and logs all raw protocol packets (>>> SENT and <<< RECV).
  • Search and filter messages by method (tools/call, initialize, etc.).
  • Protocol latency, request/response matching, and packet export.

• 🕒 Execution History & Replay:

  • Stores recent invocations with timestamps, status, input parameters, and duration.
  • Easily copy previous requests or inspect errors.

• 🔍 Auto-Discovery & Local Server Management:

  • Automatically scans and imports configurations from:
    • Claude Desktop (claude_desktop_config.json on macOS, Windows, Linux)
    • Cursor (.cursor/mcp.json & global settings)
    • Local project mcp.json / .mcp.json
  • Register any server permanently with mcptui add <name> <command>.
  • Switch between servers on the fly with F2.

---

🤖 Headless CLI & Automation

mcptui also works as a scriptable headless CLI for automated testing and CI/CD pipelines:

# Register a server into local mcp.json permanently
mcptui add notes "python examples/sample_server.py"

# List all registered and discovered servers
mcptui servers

# Headless capabilities inspection
mcptui list notes
# Or on the fly without saving:
mcptui list --cmd "python examples/sample_server.py"

# Execute a tool headlessly and get clean JSON output
mcptui call convert_temperature '{"value": 100, "from_scale": "C", "to_scale": "F"}' --server notes

# Remove a server from local config
mcptui remove notes

---

⌨️ Keybindings

Key	Action
1 – 5	Switch active tab (Tools, Resources, Prompts, Logs, History)
Ctrl + Enter	Execute active tool / Read resource / Get prompt
F2	Open Server Switcher modal
F5 / Ctrl + R	Reload server capabilities
Tab / Shift + Tab	Navigate between sidebar, editor, and response view
Ctrl + Q / q	Quit mcptui

---

📄 License

MIT License. Contributions, feedback, and stars are warmly welcome! ⭐