Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/Mats2208/MCP-Packet-Tracer/llms.txt

Use this file to discover all available pages before exploring further.

MCP Packet Tracer exposes two transport modes for connecting AI clients: stdio, where the client spawns and owns the server process, and streamable-HTTP, where the server runs independently and any number of clients connect to it over http://127.0.0.1:39000/mcp. Both modes give you full access to all 33 MCP tools and 5 resources; the difference is entirely in lifecycle, persistence, and multi-client sharing.

​
Comparison

stdiostreamable-HTTP
Who starts the serverThe MCP client spawns it automaticallyYou start it manually with python -m packet_tracer_mcp
PersistenceLives only for the client session; exits when the client closesStays running independently of any client
Multi-client supportOne client per server processMultiple clients (e.g., multiple VS Code windows) share one instance
Shared stateIsolated per client sessionShared across all connected clients
DebuggabilityLogs go to the client’s output panelYou can tail logs in the terminal; curl http://127.0.0.1:39000/mcp works
Best forVS Code, Claude Desktop, Cursor, Codex CLI — any single desktop clientMulti-window VS Code, remote setups, or when you want the server always on

​
Starting Each Mode

​
stdio

Pass the --stdio flag. The MCP client handles process spawning — you configure this in the client’s MCP config file and never run the command yourself. 
python -m packet_tracer_mcp --stdio
The server reads MCP messages from stdin and writes responses to stdout. The internal HTTP bridge to Packet Tracer (port 54321) still starts automatically inside the spawned process, so live deploy works identically in stdio mode.

​
streamable-HTTP

Omit the --stdio flag (HTTP is the default). Run this once in any terminal before opening your AI client: 
python -m packet_tracer_mcp
# Serving on http://127.0.0.1:39000/mcp
The server binds to 127.0.0.1:39000 and exposes the MCP endpoint at /mcp. It also starts the bridge on 54321. Both services stay alive until you terminate the process.

​
Client Configuration Examples

​
stdio — VS Code .vscode/mcp.json

{
  "servers": {
    "packet-tracer": {
      "type": "stdio",
      "command": "python",
      "args": ["-m", "packet_tracer_mcp", "--stdio"]
    }
  }
}

​
stdio — Claude Desktop claude_desktop_config.json

{
  "mcpServers": {
    "packet-tracer": {
      "command": "python",
      "args": ["-m", "packet_tracer_mcp", "--stdio"]
    }
  }
}

​
stdio — Cursor .cursor/mcp.json

{
  "mcpServers": {
    "packet-tracer": {
      "command": "python",
      "args": ["-m", "packet_tracer_mcp", "--stdio"]
    }
  }
}

​
stdio — Codex CLI ~/.codex/config.toml

[mcp_servers.packet-tracer]
command = "python"
args = ["-m", "packet_tracer_mcp", "--stdio"]

​
streamable-HTTP — VS Code .vscode/mcp.json

Start the server first, then configure VS Code to point at it: 
python -m packet_tracer_mcp

{
  "servers": {
    "packet-tracer": {
      "url": "http://127.0.0.1:39000/mcp"
    }
  }
}

​
When to Use Each

Use stdio when…

  • You use a single desktop AI client (VS Code, Claude Desktop, Cursor, Codex CLI)
  • You want zero-configuration — no background process to manage
  • You prefer the server to start and stop with your editor session
  • You are setting up MCP Packet Tracer for the first time

Use streamable-HTTP when…

  • You run multiple VS Code windows simultaneously and want shared topology state
  • You need the server always available, independent of any client session
  • You want to debug tool calls with curl or inspect live logs in a terminal
  • You are integrating MCP Packet Tracer into a remote or shared development environment
Even in stdio mode, the internal HTTP bridge to Packet Tracer starts automatically on port 54321 inside the spawned process. Live Deploy — sending commands directly to a running PT instance — works exactly the same in both transport modes. The choice of stdio vs. streamable-HTTP only affects how your AI client talks to the MCP server, not how the MCP server talks to Packet Tracer.

Build docs developers (and LLMs) love

Get started for freeTalk to us