Packet Tracer MCP

Cable two existing devices together. Cable defaults to copper straight-through; use 'cross' for switch trunks, 'serial' for WAN P2P (requires HWIC-2T module on ISR routers — add it first with pt_add_module), 'fiber', 'console' or 'coaxial' as needed. Convention: 'straight' for LAN segments (router↔switch, switch↔PC); 'serial' for WAN exterior P2P between routers crossing a public/ISP network. SAFETY: refuses straight/cross between two routers without `confirm_internal_lan: true` to prevent the common mistake of cabling WAN exterior links with Ethernet.

Remove a device (and all of its links) from the active workspace. Self-verifies the deletion by re-snapshotting the topology.

Remove the cable attached to one specific device port. Use this when you need to re-cable without deleting either device.

Compare two snapshots (or one snapshot vs the live canvas) and describe what changed: added/removed devices and links, changed IPs, masks, link state and power state.

Render a human-readable narration of the live workspace: inventory counts, per-router subnets, dangling addresses. Read-only.

Dry-run estimator. Build the blueprint for a recipe and report what it would allocate, without touching the live canvas.

Offline generator. Builds a recipe blueprint and synthesises the IOS CLI each device would receive — without touching PT or the bridge. Useful for documentation, classroom material, or replicating the lab on real hardware. Endpoints (PCs, APs, phones) get human-readable notes instead of CLI.

Inspect a single device: model, type, power state, and live port-by-port snapshot (IP, mask, connection state).

Snapshot the live workspace and report duplicate IPs, unaddressed router uplinks, mismatched router-peer subnets, and similar findings. Read-only.

Live per-port status: link presence, isPortUp, isProtocolUp, MAC, IP/mask, wireless flag — read directly from the Port objects.

Enumerate the device models that packet-tracer-mcp knows how to instantiate, with their port lists and the alias shortcuts accepted by other tools.

Enumerate hardware modules. Without `device`: PT's full catalog (filterable). With `device`: the modules physically installed in that chassis right now (slotPath → name).

Enumerate the topology recipes pt_cook_topology can build, with their parameter shape.

List every persisted canvas snapshot, sorted from newest to oldest.

Load a saved snapshot and diff it against the live workspace. Reports added/removed/changed devices, ports, and links since the snapshot was taken.

Inspect and prune logical clusters on the canvas. `list` returns the cluster tree; `remove` deletes a cluster by id; `uncluster` dissolves it while keeping its members. Note: programmatic cluster *creation* is not reachable from PT 9's IPC (requires UI selection).

Inspect the live workspace and apply safe, conservative repairs (e.g. powering on devices that have active links). Reports remaining issues that need a human decision.

Reposition a device on the logical canvas via Device.moveToLocation(x, y).

Open a .pkt file into the active PT workspace via AppWindow.fileOpen. PT's native fileOpen MERGES the file with the existing canvas; pass replace=true (default) to wipe the canvas first via fileNew(false). PT 9 encrypts .pkt with per-file key derivation so the leading bytes are not a stable magic; validity is decided by FileOpenReturnValue=0 from PT. Returns deviceCount before/after.

Open a .pkt provided as base64 bytes (no permanent file required on disk). The server asks PT to write the bytes to a temp path inside its user folder via SystemFileManager.writeBinaryToFile, then triggers fileNew(false)+fileOpen and removes the temp. PT 9 encrypts .pkt with per-file key derivation so there is no stable magic to validate server-side; validity is decided by FileOpenReturnValue=0 from PT. Useful when bytes come from a remote source (DB, MCP client) and the server filesystem is not visible to PT.

Run `ping <target>` on a device (PC or router) and return the parsed result: sent/received/lost/success rate plus the raw transcript. Auto-detects PC vs router output format.

Validate a topology plan BEFORE touching the canvas. The LLM declares devices, links and LAN roles; the tool returns a structured review with errors (will definitely fail), warnings (likely wrong) and a human-readable summary. MANDATORY for any topology with ≥3 routers or whenever the user supplies a diagram. Present the output VERBATIM to the user and wait for confirmation before calling pt_add_device. This is the single biggest leverage to avoid the common mistakes (Gigabit on WAN exterior, PCs on transit LANs, ad-hoc layout).

Snapshot the live topology in Packet Tracer: returns name, model, classname and coordinates for every user-placed device.

Read the live ACLs of a router via AclProcess: per-ACL canonical commands as PT would render them in 'show running-config | section access-list'.

Read NetworkFile metadata of the .pkt currently open in PT: description, file version, and the on-disk filename if it has been saved.

Read the live VLAN database of a switch via VlanManager (id, name, default flag, MAC table size).

Rename a device. Refuses if the new name is already taken so you don't accidentally collapse two devices into one.

Send a CLI command to a router or switch and return the slice of console output produced by it. Uses the device's getCommandLine().enterCommand() entry point.

Push a multi-line CLI block (e.g. an OSPF/EIGRP stanza or a sequence of `ip nat` rules) to a device. Returns a structured payload with the count of commands sent, a truncation flag, and the captured output.

Save the active PT canvas to a .pkt file at an absolute path via AppWindow.fileSaveAsNoPrompt. Returns the final size in bytes and a SHA-1 checksum for verification. Pkz format is intentionally not supported (PT 9 silently fails to write it).

Save the active PT canvas and return the .pkt as base64 bytes (no permanent file). PT writes to a temp path inside its own user folder via fileSaveAsNoPrompt, the server reads the bytes back chunked through SystemFileManager.getFileBinaryContents, then PT removes the temp. Useful when the server and PT do not share a filesystem (remote PT) or the caller wants the bytes in-memory.

Capture the current live workspace and persist it to disk under the given name. Useful for diff/audit later.

Capture the Logical workspace as a PNG/JPG file via LogicalWorkspace.getWorkspaceImage. Returns the path of the saved file. The image is base64-streamed over the bridge then decoded and written on disk by the server.

Originate a Simple PDU (ICMP echo) from one device to another via UserCreatedPDU.addSimplePdu. Returns the scenario index assigned to the PDU. Verified on PT 9 real (probe-fase8b 2026-04-29).

Escape hatch: ship arbitrary JS to PT's Script Engine. Useful when no specialised tool covers what you need yet — call e.g. 'return ipc.network().getDeviceCount();'. Prefer the typed tools when they exist.

Toggle a device's power switch (Device.setPower). Idempotent: returns 'already' when the requested state is already in effect. After power-on the boot dialog is dismissed via skipBoot.

Configure an endpoint (PC, laptop, server) for static IP or DHCP. In static mode you must pass ip+mask; gateway is optional.

Run `show ip bgp` on a router and parse the BGP table into rows (status, network, next-hop, metric, localpref, weight, path). PT 9 has no BgpProcess JS API, so this is the only way to inspect live BGP state.

Capture `show running-config` (optionally filtered by `| section`) on one or more devices. Returns a banner per device followed by its captured config slice. Useful for snapshotting state, diffing recipes, or confirming an applier landed on PT real.

Toggle PT between Realtime and Simulation modes via the RSSwitch widget. After switching to simulation, use pt_send_pdu to originate traffic and pt_simulation_play to advance.

Trigger one of the SimulationPanel buttons (play/back/forward/reset). Caller must put PT in Simulation mode first via pt_simulation_mode — calling 'play' from Realtime is a no-op.

Run `traceroute`/`tracert` on a device and return the parsed hop list. Sends both command names so it works on PCs (tracert) and IOS routers (traceroute); the wrong one is rejected by the parser and ignored.