war-devtools-mcp

The developer tools of a running WarClient64 as an MCP server on stdio, for developing WAR interface addons with an AI assistant: evaluate Lua in the interface, reload it, read its logs, write Luau type definitions, profile it, debug it (breakpoints, stepping, the stack and its variables), look at its windows, capture what it draws, trace its events and weigh each add-on's cost a frame.

It needs Node.js 18 or newer, and has no dependencies. Nothing else to install or build.

Set it up

Claude Code on Windows (npx is a batch file there, hence the cmd /c):

claude mcp add war-devtools -- cmd /c npx -y war-devtools-mcp

Claude Code on Linux or macOS:

claude mcp add war-devtools -- npx -y war-devtools-mcp

Another MCP host:

{ "mcpServers": { "war-devtools": { "command": "npx", "args": ["-y", "war-devtools-mcp"] } } }

Turn the client's endpoint on

The bridge talks to a client that is already running and has its endpoint on. Set WAR_DEVTOOLS=1 in the environment of WARDEV-64.exe or WARTEST-64.exe before the launcher starts it, or run the test client with warclient.exe --devtools. WAR-64.exe, the live build, has no endpoint.

The bridge finds the client by itself, on its first call and again after the client restarts, and tells you when none runs. Everything stays on this machine: it connects only to 127.0.0.1, with a token from a file in your own folder.

Options

| Option | What it does | |---|---| | --pid=<pid> | talk to this client, when several run; without it, the newest | | --discovery-dir=<folder> | where the clients' discovery files are, if not the usual folder |

Tools

client_status, lua_eval, reload_ui, read_logs, write_types, profile_lua, ui_windows, ui_window, ui_window_at, ui_capture, trace_events, addon_cost, debug_breakpoints, debug_wait, debug_step, debug_evaluate, debug_variables, debug_detach. Your MCP host shows what each one does and takes.

A tool's path is relative to the folder the host started the bridge in.

Licence

MIT.