Quick StartIntegrations › Claude Code

Annex — Claude Code

Step-by-step setup for Anthropic's CLI with KaKeKiKoKu via MCP

This guide walks you through configuring Claude Code so it works with KaKeKiKoKu.

Claude Code is Anthropic's command-line tool for AI-assisted development. It is the most complete integration with KaKeKiKoKu today thanks to channels: a feature that lets KaKeKiKoKu send messages to the AI in real time (automatic sync requests, drift alerts, "Talk to AI" buttons). Other AI clients work in Precise mode, copying prompts to the clipboard for you to paste — but with Claude Code the integration is bidirectional and seamless.

Important: Claude Code is NOT Claude Desktop. They are different Anthropic products. Claude Code is a CLI (command line); Claude Desktop is a desktop app. When you finish setup, make sure a console window opens, not a desktop app window. If a graphical app appears, you launched the wrong product.

Requirements

Before you start the KaKeKiKoKu integration flow, you need these regardless of OS:

Plus the OS-specific tools below:

  • Git for Windows — install from git-scm.com/install. If you install from a user account without administrator rights, the installer will request UAC elevation.
  • Git — preinstalled on most Linux distributions. If missing, install with your package manager (Ubuntu/Debian: sudo apt install git, Fedora/RHEL: sudo dnf install git, Alpine: sudo apk add git).
  • curl — used by the Claude Code installer below. Pre-installed on most desktop distros, but minimal Ubuntu/Debian and Alpine images don't include it. Install if missing:
    • Ubuntu/Debian: sudo apt install curl
    • Fedora/RHEL: sudo dnf install curl
    • Alpine: sudo apk add curl

Install or update Claude Code

If Claude Code is already installed, verify the version before continuing. If it is missing, use the OS-specific installer below.

  1. Open PowerShell (search for it in the Start menu).
  2. Paste and run Anthropic's official installer:
    powershell -ExecutionPolicy Bypass -Command "irm https://claude.ai/install.ps1 | iex"
  3. Close PowerShell completely and reopen it. This is mandatory: the new PATH only applies to processes launched after the change.
  4. Verify the installation:
    claude --version
  5. If claude --version doesn't respond, some Claude Code installs don't add .local\bin to the user PATH correctly. Paste this in PowerShell, close it and reopen:
    [Environment]::SetEnvironmentVariable("Path", [Environment]::GetEnvironmentVariable("Path","User") + ";$env:USERPROFILE\.local\bin", "User")
  6. Note the executable path. The typical Windows location is:
    C:\Users\<your-user>\.local\bin\claude.exe

    If you installed via npm instead of the native installer, it might also live at %APPDATA%\npm\claude.cmd. Anthropic now recommends the native binary for most users.

  1. Install Claude Code using Anthropic's native installer (recommended):
    curl -fsSL https://claude.ai/install.sh | bash

    This downloads a self-contained binary — you do not need Node.js or npm.

  2. Verify the installation:
    claude --version
  3. The first time you run claude, it will prompt you to log in with your Anthropic account.

Other Linux installation methods. Anthropic also publishes signed APT (Debian/Ubuntu), DNF (Fedora/RHEL), and APK (Alpine) repositories at downloads.claude.ai — useful for managed fleets or distros where you prefer your package manager. See the official installation docs for repository details and GPG verification.

Keep Claude Code up to date. Native installations update in the background. Run claude update when you want an immediate update, or choose the stable or latest channel in /config. Package-manager installs such as WinGet, Homebrew, APT, DNF, and APK use their own update flow.

↑ Back to top

1. Register KaKeKiKoKu as an MCP server in Claude Code

Register KaKeKiKoKu with Claude Code's built-in MCP command. Choose your operating system and run the command shown there:

claude mcp add --transport stdio --scope user wuniq -- wuniq-mcp.exe

Important. Use the alias name wuniq-mcp.exe. Never substitute an absolute path to a binary under C:\Program Files\WindowsApps\... — Windows ACLs cause Permission denied errors at launch.

claude mcp add --transport stdio --scope user wuniq -- /snap/bin/wuniq.wuniq-mcp

This edits your Claude Code config file automatically and leaves any other registered MCP servers intact.

The --scope user flag matters. It makes the KaKeKiKoKu registration visible from any folder. Without it, the entry stays local to the folder where you ran the command — Claude Code won't find KaKeKiKoKu if you start it elsewhere.

Verify the registration

claude mcp list

It should show wuniq among the configured servers.

If you later uninstall KaKeKiKoKu

Only if you decide to stop using KaKeKiKoKu with Claude Code, remove the MCP registration with:

claude mcp remove wuniq

Where this is stored. User-scoped servers live in ~/.claude.json and load in every project for that OS user. Claude Code also supports local scope (private to one project, stored under that project's entry in ~/.claude.json) and project scope (shared through .mcp.json). KaKeKiKoKu uses user scope here because one local KaKeKiKoKu engine can serve several projects.

↑ Back to top

2. Startup instructions (the second piece of the integration)

This is the second of the two pieces every integration needs (the first was registering KaKeKiKoKu as an MCP server in step 1). A small file in your working directory tells Claude Code to use KaKeKiKoKu from the start of every session — without it, Claude Code starts "neutral" and you have to ask for open list manually each conversation.

  1. In your working directory (the base folder you configured in KaKeKiKoKu), create this structure if it doesn't exist:
    .claude/rules/

    The path is the same on Windows and Linux — it's relative to the working directory, not to your home folder.

  2. Inside, create a file called wuniq.md with this content:
    # KaKeKiKoKu Studio (Knowledge Engine)
    
    At the start of every session, run `open list` using the KaKeKiKoKu MCP tool. If it returns projects, `open` each one. If the list is empty, proceed normally — the user may open projects later from the KaKeKiKoKu UI. Never assume the working directory is a KaKeKiKoKu project; the working directory may be a container folder holding multiple projects. Only open paths that `open list` explicitly returns.
    
    Before working on any code file, `read` its KaKeKiKoKu sidecar for context. After changing code, run `sync`.
  3. Save. Claude Code automatically loads every file in .claude/rules/*.md at the start of each session.

Why don't we put this inside CLAUDE.md?

Your CLAUDE.md (if you have one) is yours. KaKeKiKoKu prefers not to touch it so it doesn't overwrite content you wrote. .claude/rules/wuniq.md is an independent file KaKeKiKoKu controls and you can delete freely without losing anything of your own.

If this repository already uses AGENTS.md. Claude Code does not load AGENTS.md directly. Keep the dedicated KaKeKiKoKu rule above, or create a CLAUDE.md containing @AGENTS.md so Claude can import shared cross-client instructions without duplicating them. On Windows, the import is simpler than a symlink.

What if I skip this step?

KaKeKiKoKu still works, but Claude Code won't open your projects automatically. You'll have to ask it to run open list the first time in every conversation. If the list is empty, open the project manually from KaKeKiKoKu or tell Claude Code the project path.

↑ Back to top

3. Launch Claude Code

Open a terminal in your working directory and run:

claude --dangerously-load-development-channels server:wuniq

What this flag does

--dangerously-load-development-channels server:wuniq opts the explicitly named KaKeKiKoKu MCP server into Claude Code's channels research preview. Channels let KaKeKiKoKu push messages into the running session. Without the flag KaKeKiKoKu still works, but in Precise mode: sync is not pushed automatically and "Talk to AI" buttons copy to the clipboard.

About the flag's name

The flag includes dangerously because it bypasses Anthropic's research-preview allowlist for the specific development channel named after server:. It does not mean "disable every permission", but it should still be used only with a local server you trust. KaKeKiKoKu names the exact server instead of enabling unknown channels. If you prefer not to use a preview feature, see "No-channels mode" below.

Organization policy. Personal Pro and Max accounts can opt in per session. Team and Enterprise organizations must also enable Channels in their Claude Code admin settings; otherwise KaKeKiKoKu's MCP tools still work, but pushed channel messages do not arrive.

Useful options

Shortcut: create your own launcher (optional)

To avoid typing the command every time, you can create a launcher script in your working directory. KaKeKiKoKu does not generate it for you — you control the flags and any extra steps before or after.

Create a file called KaKeKiKoKu Run.cmd (or any name you prefer) in your working directory with this content:

@echo off
set "PATH=%USERPROFILE%\.local\bin;%PATH%"
cd /d "%~dp0"
powershell.exe -NoExit -Command "claude --dangerously-load-development-channels server:wuniq"

Double-click and Claude Code launches ready to work. You can pin it to the desktop or taskbar. Adjust flags as needed: add --continue to resume the previous session, or --model opus[1m] to request Opus with extended context (availability depends on your plan) — see "Plan and model" below.

Create a file called wuniq-run.sh (or any name you prefer) in your working directory with this content:

#!/bin/bash
export PATH="$HOME/.local/bin:$PATH"
cd -- "$(dirname -- "$0")"
claude --dangerously-load-development-channels server:wuniq

From the folder where you saved it, make it executable once:

chmod +x wuniq-run.sh

Run it from the terminal with ./wuniq-run.sh. Some desktop environments can also run .sh files on double-click, but terminal launch is the predictable option. Adjust flags as needed: add --continue to resume the previous session, or --model opus[1m] to request Opus with extended context (availability depends on your plan) — see "Plan and model" below.

What to expect on startup

↑ Back to top

4. First launch

The first time Claude Code connects to KaKeKiKoKu you'll see a series of prompts. Here's what to expect and what to choose.

Folder permission

Claude Code asks if you trust the working directory. Choose "Always allow".

MCP server permission

Claude Code may ask for additional confirmation to use the wuniq.run tool. Accept. That's what allows the AI to read and write context through KaKeKiKoKu.

Anthropic login

If this is your first Claude Code, complete the login with your Anthropic account. KaKeKiKoKu does not participate in this step — subscription and authentication are managed entirely by Anthropic.

Console window

Make sure a terminal/console window opens where you talk to the AI. On Windows, if a graphical desktop app opens instead of a console, you launched Claude Desktop by mistake (different product). Close it and review the command.

Verify KaKeKiKoKu is connected

Inside Claude Code, type:

/mcp

It should list wuniq among the available servers. If it doesn't appear, Claude Code didn't find the registration — go back to step 1.

First use

If you followed step 2 (.claude/rules/wuniq.md), Claude Code will run open list automatically when you start your first conversation.

If you didn't create the file, ask Claude Code to run open list. If the list is empty, open the project manually from KaKeKiKoKu or tell Claude Code the project path.

If something fails after login

Close Claude Code, go back to the console, and relaunch. If it still doesn't work, see "Troubleshooting" at the end.

↑ Back to top

5. Plan and model

KaKeKiKoKu works through your Claude Code subscription (it does not use the Anthropic API). The Anthropic plan you have determines which model is the default and how much context the AI can handle in a single session.

Default model by plan

Claude Code's default selection follows your account type and any organization policy. At the time of this review, Anthropic documents:

If you launch claude --dangerously-load-development-channels server:wuniq without --model, Claude Code uses your plan's default. Perfectly valid for most sessions.

1M-token extended context

Claude Code now supports a 1M-token context window on Fable 5, Sonnet 5, Opus 4.6 and later, and Sonnet 4.6. The exact behavior depends on the model, provider, and plan:

Plan Opus + 1M context Sonnet + 1M context
Max, Team, Enterprise Automatically upgraded to 1M Sonnet 5 is native 1M on Anthropic; Sonnet 4.6 requires usage credits
Pro Requires usage credits for an Opus 1M variant Sonnet 5 is native 1M on Anthropic; older 1M variants can require credits

Usage-credit and provider rules can change independently of KaKeKiKoKu. See usage credits in Anthropic's support docs and the current model configuration page before relying on a particular entitlement.

Selecting a model

Anthropic provides model aliases so you don't have to remember exact version numbers:

Examples (combine with the launch flag):

claude --model best --dangerously-load-development-channels server:wuniq
claude --model opus[1m] --dangerously-load-development-channels server:wuniq
claude --model opusplan --dangerously-load-development-channels server:wuniq

You can also pin a specific version with the full model name, but aliases are recommended — they keep pointing to the current version as Anthropic releases new models.

Working with a smaller context window

If your selected model, provider, or organization caps the session below 1M, KaKeKiKoKu remains productive because context loading is progressive: it does not send the entire project as raw text. Claude reads sidecars and uses search / read on demand.

Up to date

Anthropic releases new models periodically. Check the Claude Code model configuration docs for current aliases, default behavior, and version-pinning options.

↑ Back to top

6. No-channels mode

If you'd rather not use the --dangerously-load-development-channels flag (because the name bothers you, because you don't need push notifications, or for any other reason), you can launch Claude Code like this:

claude

KaKeKiKoKu still works — but in Precise mode instead of Automatic.

What changes in Precise mode

When to use this mode

KaKeKiKoKu configuration

In KaKeKiKoKu > Settings > Application > AI Sync, choose Precise mode. That replaces automatic push with clipboard copy for all affected functions.

↑ Back to top

7. Troubleshooting

Common problems and how to solve them.

claude --version doesn't respond after install

Some Claude Code installs don't add .local\bin to the user PATH correctly. Paste this in PowerShell, close it and reopen:

[Environment]::SetEnvironmentVariable("Path", [Environment]::GetEnvironmentVariable("Path","User") + ";$env:USERPROFILE\.local\bin", "User")

After installing Claude Code with curl install.sh, the updated PATH only applies to new shells. Close the terminal where you ran the installer and open a new one. If claude --version still fails, source your shell config in the active terminal:

source ~/.bashrc   # or ~/.zshrc, depending on your shell

KaKeKiKoKu doesn't appear in /mcp

  1. From a terminal, run claude mcp list. If wuniq is missing, repeat step 1.
  2. If wuniq is listed but still does not load, remove the registration and add it again:
    claude mcp remove wuniq
  3. When you repeat step 1, make sure you use the command for your OS:
    wuniq-mcp.exe
    /snap/bin/wuniq.wuniq-mcp
  4. Close Claude Code completely and relaunch. The MCP registration is loaded at startup.

"Permission denied" or "EPERM" error launching Claude Code (Windows)

If a previous registration points to the KaKeKiKoKu executable using an absolute path under C:\Program Files\WindowsApps\..., Windows refuses execution due to restrictive ACLs. Remove the registration and repeat step 1 using the alias wuniq-mcp.exe:

claude mcp remove wuniq

Claude Code launches but KaKeKiKoKu doesn't respond

Channels don't work ("Talk to AI" buttons don't arrive)

The --model flag fails

Check claude --help or the Claude Code model configuration docs. Some current models require a newer Claude Code build; update first if aliases such as fable, best, or a 1M variant do not appear.

Reinstalling KaKeKiKoKu leaves a stale entry in the config file

If you reinstalled KaKeKiKoKu and the connection fails, remove the old Claude Code registration and repeat step 1:

claude mcp remove wuniq

KaKeKiKoKu can't see projects outside /home/ (Linux)

The KaKeKiKoKu snap is sandboxed by default and cannot read files outside your home folder unless you grant that permission. If your projects live in /mnt/, /media/, a USB drive, or a VM shared folder such as /mnt/hgfs/, run:

sudo snap connect wuniq:removable-media

Then reload the KaKeKiKoKu web UI and open the project again.

Very large projects stop updating automatically (Linux)

Linux caps the number of files a single user can watch at once. Very large projects can hit that limit. Symptoms: file changes go undetected, KaKeKiKoKu panel doesn't update. Raise the watch limit:

echo "fs.inotify.max_user_watches=524288" | sudo tee -a /etc/sysctl.conf
sudo sysctl -p

Closing Claude Code also closes KaKeKiKoKu (Linux)

On Linux, if Claude Code started KaKeKiKoKu through the MCP connection, KaKeKiKoKu may close when Claude Code closes. If you want KaKeKiKoKu to stay open between Claude Code restarts, launch KaKeKiKoKu from your desktop menu before launching Claude Code.

File changes go undetected on shared/network mounts (Linux)

Some shared folders and network drives do not notify Linux apps reliably when files change. If your project is on a VMware shared folder, SMB/CIFS share, or sshfs mount, KaKeKiKoKu may not notice edits immediately even after access has been granted. When that happens, run sync from Claude Code or click Analyze in the KaKeKiKoKu UI.

Still stuck?

Email support@kakekikoku.com with your operating system, Claude Code version (claude --version) and the contents of your Claude Code config file (Windows: %USERPROFILE%\.claude.json, Linux: ~/.claude.json). KaKeKiKoKu can also generate a diagnostic report (Support function in the app) that captures the infrastructure state — attach it to the email.

↑ Back to top