Local AI Automation

Since 2.3, WPE can hand its existing features to an AI client on the same machine that speaks MCP: read packets, look at logs, write filters, create accounts, run senders and robots — all by conversation. This chapter covers the connection path, client setup, practical workflows, and WPE's switches and audit trail.

01 · What this is

MCP (Model Context Protocol) is a common interface between AI clients and programs on the same machine. WPE 2.3 implements the server side: it wraps its own features into named "tools" that a connected AI client can call — so instead of clicking through menus you can say "change the third byte of that packet to 0x10 and send it again".

• It is local: the service runs alongside WPE on this computer and serves the current Windows user only. No internet, no account.
• WPE has to be running: the connector presses WPE's buttons for you; it never starts anything. Close WPE and the AI client reports the service offline.
• WPE is the only source of truth: every read and write goes through WPE's own entry points — the AI client never edits your configuration files or database behind its back.
• Who uses it: you. Whoever runs WPE can drive it over MCP, so this has to be your own, trusted machine.

02 · MCP Settings

The entry is the second narrow row on the home screen (the one labelled MCP), with the number of available tools on the right. It opens a sectioned dialog like every other setting: two switches and one list.

The MCP tool list below shows every tool actually registered, grouped as proxy / inject / packet / proxy accounts / filter / send / robot / warehouse / decoders / extractors / settings / other, each with its name and a one-line description — this list matches what the AI client sees one for one. Write tools are marked according to the confirmation switch above. This tutorial also carries the list on the web — see Chapter 9 · MCP Tool Reference.

03 · State and logs

• The MCP status lamp (right-hand end of the status bar, right after the proxy state): grey = MCP service off; amber = running, writes need your confirmation; green = running, writes can execute automatically. One lamp tells you which state you are in
• The MCP log (Proxy / Inject Mode → System Log, the fourth tab — see Tools, section 07): every tool call, the write-audit verdict and any error. The module column carries the tool name, so this is the page to read when you want to know what the AI actually touched

04 · Connecting an AI client

The WPE release ships a self-contained connector, WPEMcpServer.exe (no extra runtime needed). The launcher syncs it to a fixed path on every start:

C:\WPE64DB\McpServer\WPEMcpServer.exe

1. Add a stdio server to your AI client's MCP configuration with that path as the command (if the client wants a full command line, this exe on its own, with no arguments).
2. Keep WPE running and let the client connect once: it discovers the WPE instance running under the current user and pulls the tool list.
3. Close WPE and the client reports the service offline; connect again next time WPE is up. Nothing in the configuration changes.

Using VS Code

VS Code puts MCP inside GitHub Copilot Chat — MCP servers are used through Copilot's agent mode, so install the Copilot extension and sign in first (with Copilot unavailable, the server in your configuration will not start), then connect WPE in three steps.

1. Open the configuration file: run MCP: Open User Configuration from the Command Palette (Ctrl+Shift+P) — that is the user-level file, configured once and available in every workspace; to scope it to one project, open that project's .vscode/mcp.json. You can also run MCP: Add Server and pick Workspace or Global, which writes the block below for you.
2. Paste this in. Note that backslashes in a Windows path must be doubled; the file has completion and validation, so a misspelled key is flagged as you type:
   mcp.jsonCOPY

   { "servers": { "wpe": { "type": "stdio", "command": "C:\\WPE64DB\\McpServer\\WPEMcpServer.exe" } }}
3. Start it and trust it: once saved, a Start link appears in the file — click it. The first start asks you to confirm that you trust the server; once confirmed, VS Code discovers the tools and exposes them in chat. If the tool list goes stale (after a WPE upgrade, say), run MCP: Reset Cached Tools.

Then open the Chat view (Ctrl+Alt+I), switch to agent mode and just ask in plain language — "read me the current packet list", for instance. The Configure Tools button next to the chat input lists every WPE tool and lets you enable or disable them one by one.

Using Trae

Trae can use WPE through an MCP configuration in your project too. Keep WPE running first; create a .vscode folder at the project root, then add mcp.json with the following content. Backslashes in a Windows path must be doubled:

Save the file, then start or refresh the wpe service in Trae's AI / MCP management view and ask directly — for example, “show the current capture state” or “list the latest 20 packets but do not modify anything”. WPE's connector already exposes MCP over stdio, so do not add --mcp, --pipe, or --mcp-full arguments found in examples for other products; it automatically discovers the WPE running for the current Windows user.

Using Claude Code

Claude Code can use WPE as a local stdio MCP server too. Keep WPE running, then execute this command in a terminal; the quoted path prevents the shell from splitting a path containing spaces or special characters:

Run claude mcp list afterwards to check that wpe appears, then ask Claude Code directly. If it offers a project-level or user-level scope, choose project-level for one project only or user-level for all your projects. The path survives WPE upgrades; if the tool count or descriptions still look old, reconnect the service so the client refreshes its tool list.

05 · What it can do

Tools come in read-only and write flavours. Read tools let the AI look; write tools let it change things — the latter are governed by "MCP operations need confirmation".

• One extra thing on the start page: while you are still on the home screen and have not entered a mode, the AI can pick Proxy or Inject Mode for you. That is only a choice between the two modes — it starts no listener and injects nothing — and the ability disappears once you leave the home screen (picking the mode you are already in just reports that nothing needs doing)
• Two things are not offered: arbitrary file and clipboard access, and injection that runs without confirmation

06 · A steadier way to work with AI

MCP works especially well when one conversation follows “inspect → propose → apply → verify”. Tell the AI both your goal and your boundaries, then let the tools follow that order instead of turning a guess straight into a write.

07 · Using it, and troubleshooting

MCP hands WPE's controls to an AI, so keep these in mind:

• Only enable it on your own computer: the service and its discovery record are visible to the current Windows user only, so other users on the same machine cannot see them. Turning the master switch off when you are done is the simplest habit
• When in doubt, keep confirmation on: with it on, every write asks inside WPE — the decision to execute stays with you
• Read tools return complete data: packet contents and account passwords are handed to the client as they are. Connect only clients you trust