JidoDebugger by dhq-boiler
// getting started

How to use JidoDebugger

From install to your first AI-driven debugging session, in four steps.

JidoDebugger ships a local MCP server (jidodebugger-mcp) that runs on your own Windows machine. You connect it to the AI agent you already use — Claude Code, Claude Desktop, or Cursor — and then just ask the agent to run, watch, and debug your app in plain language.

  1. 01

    Install JidoDebugger

    Claim a free license and download the installer from the home page, then run it. The installer puts the local MCP server and the jidodebugger CLI on your machine (Windows 11, x64).

  2. 02

    Connect your AI agent

    Register the local MCP server with your AI client. Pick your client below and run the command in any terminal — the installed server path is resolved automatically, so there is nothing to hand-edit.

  3. 03

    Restart your AI client

    Fully quit and reopen the client so it picks up the new MCP server. The JidoDebugger tools then show up in its tool list.

  4. 04

    Ask it to debug

    That's it. Talk to your agent in plain language — it drives your app through JidoDebugger's tools and shows you what it actually sees on screen.

Step 2 — run this in any terminal after installing:

jidodebugger register --client claude-code
Writes to: %USERPROFILE%\.claude\claude.json

Server not found? Install JidoDebugger first, or set the JIDODEBUGGER_MCP_PATH environment variable to the full path of jidodebugger-mcp.exe. You can check the resolved path anytime with: jidodebugger which-mcp

Then try a prompt like

  • Launch Notepad, type "hello", and confirm the title bar updates.
  • Attach to my running app, open the Settings dialog, and tell me if any control is clipped or cut off.
  • Run my CLI tool with --help and check the exit code is 0.
  • Watch my app for 30 seconds and report any crash, hang, or error dialog.

Good to know: your AI agent reads screenshots itself — it calls capture and reasons over the image directly, so a multimodal model like Claude needs no extra vision setup for visual checks. The optional JIDODEBUGGER_VLM_PROVIDER (ollama / anthropic / openai / gemini) only powers two server-side tools — assert_visual_state and intent — handy for "wait until this state holds" polling assertions or non-multimodal agents. UIA, clicking, typing, capture, and OCR never need it.