Documentation
Everything you need to install, connect your MCP client, and deploy your first topology. No prior MCP experience required.
Claude Code, OpenCode, Codex, Cursor — any MCP client. It does the install for you.
Tested on 9.0 and 9.0.1, on Windows and macOS. Free with a NetAcad account.
That's all. Python comes with uv, which the agent installs if it's missing — no PATH to fix, no pip.
Install the Packet Tracer MCP from https://github.com/Mats2208/MCP-Packet-Tracer and set everything up for me.
It reads the repository, follows llms-install.md, registers the server, installs the skill and puts the extension where Packet Tracer finds it.
Open Packet Tracer. It says it "has detected a new script module" — click Yes. Restart your agent and ask it to build a network.
# 1. uv (brings its own Python) curl -LsSf https://astral.sh/uv/install.sh | sh # macOS · Linux powershell -c "irm https://astral.sh/uv/install.ps1 | iex" # Windows # 2. register the server (Claude Code; in PowerShell quote the "--") claude mcp add --scope user --transport stdio packet-tracer -- uvx packet-tracer-mcp --stdio # 3. the skill, so the agent discovers instead of guessing uvx packet-tracer-mcp install-skill
Then add the extension: see Live Deploy.
claude mcp add (or any stdio client config), the client spawns
uvx packet-tracer-mcp --stdio for you. The internal HTTP bridge to Packet Tracer
(:54321) still starts automatically inside that process, so live deploy works the same. Want one shared instance for multiple clients? Run it yourself in streamable-HTTP mode:
$ uvx packet-tracer-mcp INFO Packet Tracer MCP v0.12.0 INFO Transport: streamable-http INFO Listening on http://127.0.0.1:39400/mcp
Then point any client at http://127.0.0.1:39400/mcp.
Add to .vscode/mcp.json:
{
"servers": {
"packet-tracer": {
"type": "stdio",
"command": "uvx",
"args": ["packet-tracer-mcp", "--stdio"]
}
}
} Add to claude_desktop_config.json:
{
"mcpServers": {
"packet-tracer": {
"command": "uvx",
"args": ["packet-tracer-mcp", "--stdio"]
}
}
} uvx packet-tracer-mcp --stdio) — recommended for desktop clients —
or point an HTTP client at http://127.0.0.1:39400/mcp. The server exposes 79 tools and its resources that any
MCP-compatible client can discover automatically. If your client can't find uvx, restart it after installing uv, or use the full path: ~/.local/bin/uvx (macOS/Linux), %USERPROFILE%\.local\bin\uvx.exe (Windows).
Once connected, ask your AI agent to build a topology. You can also call tools directly:
pt_full_build( routers=2, pcs_per_lan=2, dhcp=True, routing="ospf", has_wan=True ) # Returns: plan + script + configs + explanation
# 1. Generate plan plan = pt_plan_topology(routers=2, pcs_per_lan=2, routing="static") # 2. Validate pt_validate_plan(plan) # 3. Auto-fix if needed pt_fix_plan(plan) # 4. Generate artifacts pt_generate_script(plan) # topology build script (JS) pt_generate_configs(plan) # IOS CLI configs # 5. Get explanation pt_explain_plan(plan) # Natural language summary
# Save everything to a directory pt_export(plan, path="./my_lab/")
.pts script module from Releases).
Live deploy streams commands into a running PT via a bidirectional HTTP bridge — no copy-paste, no bootstrap snippet. Your agent does this. By hand: download MCP-Control-Center.pts from Releases and drop it in Packet Tracer's extensions folder, in a subfolder of its own — PT asks to enable it at the next start:
macOS Cisco Packet Tracer <version>.app/Contents/extensions/MCPControlCenter/ Windows C:\Program Files\Cisco Packet Tracer <version>\extensions\MCPControlCenter\ (admin)
Or the classic way: Extensions → Scripting → Configure PT Script Modules → Add…
This installs our own extension — you do not need any third-party tool.
Open Extensions → MCP BUILDER. The MCP Control Center window appears and auto-connects to the bridge — its polling loop is built in, so there's nothing to paste.
# Check the bridge — should report ACTIVE and CONNECTED pt_bridge_status() # Deploy a plan in real-time pt_live_deploy(plan)
Devices, links, and configurations appear in Packet Tracer within seconds. Full steps in the Live Deploy docs ↗.
Save and load topologies for later reuse.
# List saved projects
pt_list_projects()
# Load a previously saved project
pt_load_project("my_lab")
# Export creates a project automatically
pt_export(plan, path="./my_lab/") Run uvx packet-tracer-mcp --stdio in a terminal and read the error. If an old version is cached, uvx packet-tracer-mcp@latest --stdio refreshes it.
The client must find uvx: restart it after installing uv, or put the full path to uvx in the command field. For HTTP mode, verify the server is reachable on http://127.0.0.1:39400/mcp.
The MCP Control Center extension isn't registered: put the .pts back in PT's extensions folder (or add it via Extensions → Scripting → Configure PT Script Modules → Add…), then open Extensions → MCP BUILDER — it auto-connects (no snippet to paste). Make sure port 54321 is free.
Minimize the MCP Control Center window (don't just push it behind PT). When the webview is visible but in the background, Chromium keeps rendering and competes for the GPU.
Run pt_fix_plan(plan) to auto-correct common issues (wrong cables, model limitations, port conflicts). If errors persist, check the error codes for specific guidance.
Routing covers static, OSPF, EIGRP and RIP (static and OSPF are the most mature). NAT (static/dynamic/PAT) and ACLs (standard/extended/named) can be applied on live routers via the bridge. VLAN and STP automation aren't in the planner yet.
The full documentation — tool reference, device catalog, and networking guides — lives on the docs site.