MCP Integration¶

This guide covers advanced configuration for the panel-live MCP server. For a quick introduction, see the Getting Started with MCP tutorial.

Transport options¶

stdio (default)¶

panel-live mcp

Used by VS Code Copilot Chat, Claude Desktop, and Claude Code. The MCP client starts the server as a subprocess and communicates via stdin/stdout.

Streamable HTTP (for Claude.ai and remote access)¶

panel-live mcp --transport streamable-http --port 5002

Starts a Streamable HTTP server. This is the transport Claude.ai uses for remote MCP connectors. Expose it via ngrok or a public server to use from Claude.ai. The server listens on http://localhost:5002 by default.

HTTP / SSE¶

panel-live mcp --transport http --port 5002
panel-live mcp --transport sse --port 5002

Legacy SSE-based transports. Useful for testing and tools that don't yet support Streamable HTTP. Note: SSE may be deprecated by MCP clients in the future.

Configuration examples¶

Claude.ai (remote connector)¶

Claude.ai requires a publicly accessible MCP server. We hope to host a public panel-live MCP server in the future — for now, you can test with a local tunnel.

1. Start the server and expose it via ngrok:

   panel-live mcp --transport streamable-http --port 5002
   # In another terminal:
   ngrok http 5002
   

2. In Claude.ai, go to Settings > Connectors > Add custom connector.

3. Paste the ngrok URL (e.g., https://abc123.ngrok-free.app/mcp/).

Available on free plans (1 connector) and Pro/Max/Team/Enterprise.

VS Code Copilot Chat¶

Create .vscode/mcp.json in your project:

{
  "servers": {
    "panel-live": {
      "command": "uvx",
      "args": ["--from", "panel-live[mcp] @ git+https://github.com/panel-extensions/panel-live.git", "panel-live", "mcp"]
    }
  }
}

Claude Desktop¶

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "panel-live": {
      "command": "uvx",
      "args": ["--from", "panel-live[mcp] @ git+https://github.com/panel-extensions/panel-live.git", "panel-live", "mcp"]
    }
  }
}

Claude Code¶

Add to .mcp.json or your settings:

{
  "mcpServers": {
    "panel-live": {
      "command": "uvx",
      "args": ["--from", "panel-live[mcp] @ git+https://github.com/panel-extensions/panel-live.git", "panel-live", "mcp"]
    }
  }
}

Troubleshooting¶

Missing dependency¶

If you see:

Error: fastmcp is not installed. Install it with:

    pip install panel-live[mcp]

Then restart your MCP client.

Install the MCP extra and restart your client (VS Code, Claude Desktop, etc.):

pip install "panel-live[mcp] @ git+https://github.com/panel-extensions/panel-live.git"

Server not starting¶

Verify the CLI works:

panel-live mcp --help
panel-live --version

If panel-live is not found, ensure it is installed in the active Python environment and the Scripts/bin directory is on your PATH.

Slow loading in Claude.ai¶

Claude.ai does not provide COOP/COEP headers, so Pyodide falls back to a much slower initialization path (2+ minutes). VS Code Copilot Chat provides these headers and loads in ~5 seconds. This is a known limitation of the MCP Apps runtime environment.

Network errors¶

The MCP App loads assets from:

• panel-extensions.github.io — panel-live JS/CSS
• cdn.jsdelivr.net — Pyodide, Panel, Bokeh
• pypi.org / files.pythonhosted.org — Python packages
• unpkg.com — MCP Apps SDK

Ensure these domains are accessible from the client's network.

Security model¶

The MCP App runs in a sandboxed webview with a Content Security Policy (CSP) that restricts which domains can load scripts, styles, and make network requests. The CSP allows:

• 'unsafe-eval' and 'wasm-unsafe-eval' — required by Pyodide for Python-to-JS interop and WebAssembly loading
• 'unsafe-inline' — required by panel-live for inline styles
• CDN domains for Pyodide, Panel, Bokeh, and plotting libraries
• blob: and data: URIs for Pyodide workers and Bokeh images

All code execution happens in the browser via Pyodide (WebAssembly). No code is sent to any external server. The Python runtime is completely client-side.

Known limitations¶

• Server-side resources: Code runs in the browser via Pyodide. Databases, filesystem, and network APIs are not available.
• Unsupported APIs: Methods that open a GUI window (e.g. plt.show(), .plot()) do not work — return the figure object instead. Panel templates like FastListTemplate are not supported in panel-live.
• Server-based frameworks: Dash, Gradio, and Streamlit require a running server and are not supported.

FastMCP direct usage¶

The server can also be run directly with FastMCP:

fastmcp run panel_live.mcp:mcp

Or used programmatically:

from panel_live.mcp import create_mcp_server

server = create_mcp_server()
server.run(transport="stdio")