MCP Server

[leaanthony]

A place to provide feedback on the MCP Server.

[taliesin-ai]

👋 This thread is where we collect feedback on the experimental MCP Server — the built-in Model Context Protocol server that lets LLM agents (Claude Code, IDE assistants, or any MCP client) inspect, test, and drive a running Wails app. Thanks for trying it.

The MCP Server is an experiment, and what we hear here decides where it goes. The most useful feedback is concrete:

• What you ran — the wails3 command or task, your client (Claude Code, IDE, etc.), and your OS/arch.
• Which tools you used — e.g. js_eval, dom_query, screenshot_dom, mouse_click, call_bound_method, emit_event, wait_for_event.
• What you expected vs. what actually happened.
• Was it useful? Did letting an agent drive the app help you test, debug, or build something faster — or not worth the setup?
• What broke? Anything the agent got wrong: a tool that misbehaved, a snapshot that was stale, input that didn't land where expected.
• What's missing? A tool or capability you reached for that wasn't there.
• What would have to be true for you to use it regularly?

Trying it: the MCP server is compiled in only with the mcp build tag, so it's off by default and adds nothing to a normal build. Opt in with the WAILS_MCP environment variable:

WAILS_MCP=1 wails3 dev      # or: wails3 build -tags mcp

By default it listens on http://127.0.0.1:9099/mcp (streamable HTTP); set WAILS_MCP_PORT to change the port. To connect Claude Code:

claude mcp add --transport http wails http://127.0.0.1:9099/mcp

Then ask the agent to do things like "take a DOM snapshot", "click Increment three times and verify the counter", or "call the bound method with these args". There's a runnable playground at v3/examples/mcp (go run -tags mcp .) with one widget per tool category and an animated cursor so you can watch the agent drive the app.

📖 Full docs: https://v3.wails.io/guides/mcp-service/