Integrations — KaKeKiKoKu Studio Documentation

KaKeKiKoKu Studio is built for AI coding agents, such as Claude Code, Codex, Antigravity, OpenCode and the like, and for the projects you build with them. The guides on this page connect each agent to Studio.

Start quickly with one instruction file. No MCP setup required. Open Studio, add the shared instructions where your local AI client loads them, and start a new conversation. This is the recommended way to get started.

Follow the common connection guide →

The content is the same for every client. We recommend AGENTS.md as the shared file for most clients, including recent Claude Code versions. Keep existing Claude instructions or use your editor’s equivalent; the common guide explains which file loads. The AI needs tools that can reach Studio on your computer. Keep Studio running.

Your client provides access to its model, through a subscription, API credentials, or a local model. Studio does not call an AI provider itself.

MCP for everyday work

You do not need MCP for a first look at a small project; you will want it for everyday work. Use the instruction-file guide above for the quickest setup. Over HTTP the AI takes more steps, long responses can arrive trimmed, and it can use somewhat more tokens.

For everyday work, we recommend MCP. Your AI gets a dedicated KaKeKiKoKu tool: each command is a single tool call, instead of a shell command it writes out and your client may ask you to approve. Once configured, your AI client also starts Studio when the AI needs project access, so you do not have to open the application first.

MCP setup depends on your client and uses two pieces:

Point your AI to KaKeKiKoKu Studio via MCP — register KaKeKiKoKu as an MCP server in your AI client.
Add a minimal startup instruction — a small file or rule that tells your AI to use KaKeKiKoKu Studio at the start of every session.

Two complementary piecesConnect the tools. Establish the habit.

MCP connectionThe AI can reach KaKeKiKoKu.

Register the server in your AI client.

Startup instructionsThe AI knows to use it.

Add the rule for the start of each session.

A session with project contextDiscover available projects. Open the ones the task needs.

Each client guide below covers both pieces.

Most clients can share AGENTS.md, including recent Claude Code versions in the CLI and Desktop’s Code tab. Existing Claude instructions and dedicated .claude/rules/wuniq.md rules remain valid. Some clients have other recommended mechanisms, such as Antigravity 2.0’s Always On workspace rules under .agents/rules/. Each guide covers the setup, loading conditions, and verification for its client.

Model-agnostic does not mean every session feels identical. KaKeKiKoKu Studio is not tied to any specific AI model or provider. The real experience can still vary with the model you choose, the client implementation, your plan, available context, and the provider's service health at that moment. If a provider is degraded, overloaded, or a model is simply performing poorly in a given session, KaKeKiKoKu may still be configured correctly even if the interaction feels less smooth.

MCP clients and setup guides

These guides and status labels cover MCP. They reflect recorded use with specific configurations; results can vary by client version and model. Choose your client for its setup guide.

Status legend

✓ Tested configuration — a test with identified client, operating system, and KaKeKiKoKu versions
✓ Observed use — recorded use, with incomplete version or environment details
⚠ Not tested — no recorded integration test for this combination
○ No Studio build — KaKeKiKoKu has no public build for this platform

Client	Windows	macOS	Linux
Anthropic
Claude Code Anthropic's CLI — recorded Automatic-mode use with channels enabled.	✓ Observed use	○ No Studio build	✓ Observed use
Claude Desktop Anthropic's desktop app, Code tab — Precise mode. Linux client currently beta.	✓ Observed use	○ No Studio build	⚠ Not tested
OpenAI
ChatGPT Desktop (Codex) OpenAI's desktop app with local Codex projects — Precise mode. Shares MCP configuration with Codex CLI and the IDE extension.	✓ Observed use	○ No Studio build	⚠ Not tested
Codex CLI OpenAI's terminal client — Precise mode. The guide includes an optional workaround for a reported Windows `stderr` issue.	⚠ Not tested	○ No Studio build	✓ Observed use
Google
Antigravity 2.0 Google's standalone command center for agent projects — Precise mode. The guide also preserves IDE and CLI configuration paths.	✓ Observed use	○ No Studio build	✓ Observed use
Multi-model
OpenCode Open-source coding agent for many providers and models — Precise mode.	✓ Tested configuration	○ No Studio build	⚠ Not tested
Z.ai
ZCode Z.ai's agentic development environment for GLM — Precise mode. Linux client currently beta.	⚠ Not tested	○ No Studio build	⚠ Not tested
Other MCP clients
Generic MCP setup Patterns for other AI coding agents with MCP support, including Kimi Code and Qwen Code.	⚠ Not tested per client	○ No Studio build	⚠ Not tested per client

KaKeKiKoKu is available for Windows and Linux; see Download. There is no public macOS build. OpenAI also provides a Linux desktop client, but its integration with KaKeKiKoKu has not been tested here. Client availability and integration testing are separate questions.

Beyond the dedicated guides

The MCP ecosystem moves faster than this matrix. Clients without a dedicated annex may still work well — we just have not measured them, and we do not commit to documenting every client-specific bug.

Other AI coding agents — Kimi Code and Qwen Code are notable current examples. Follow the generic guide for their common MCP and startup-instruction pattern.
Models inside multi-model clients — MiniMax M3, LongCat-2.0, DeepSeek V4, MiMo V2.5, and similar models do not need separate KaKeKiKoKu setup pages. The integration belongs to the client that orchestrates them.
Unofficial third-party desktop apps (community ports, repackaged builds) — we recommend the official client from each vendor.

Found a KaKeKiKoKu Studio + client combination that works well and is not in the matrix? Tell us — we may add it to a future review.