# Karakuru MCP setup assistance

Use this guide only after the user asks to install or configure Karakuru MCP.

## Entry points

- In the startup canvas, find stable memo `guide_mcp`, normally displayed as A6 and titled “5. Claude／Codexから操作する”. Use its download button.
- Direct package URL: `/karakuru-mcp/Karakuru-MCP-Setup-0.9.6.zip`.
- Browser-only setup does not require a Karakuru data-folder path. Configure a Union or Cluster root only when direct folder access is needed.
- After extraction, read `START_HERE.md`; it is the authoritative platform-specific installer guide included with the package.

## Ask before changing anything

1. Is the target Claude Desktop, Claude Code/Codex, or both?
2. What absolute folder should contain Karakuru user data?
3. May an existing `karakuru` MCP configuration or skill be replaced?
4. Should Jev semantic search be configured? If yes, let the user provide the key through the supported local configuration flow; do not echo it unnecessarily.

Do not choose the Karakuru application source directory as the user's data folder without explicit direction. A dedicated folder such as `Karakuru-Data/Union` or `project/notes/karakuru` is preferred.

## Verification

1. Restart the AI client when its configuration requires it.
2. Confirm that the `karakuru` MCP server and tools are visible.
3. Run `list_cells`, then `read_cell`, as read-only checks.
4. For live browser notes, open Karakuru and have the user enable “Claude・Codexとの接続”, then call `list_browser_sessions` or `list_browser_cells`.
5. Perform a write test only when the user asks for it. Confirm that Ctrl+Z restores browser changes.
