Connect your AI assistant

Connect Maru to Claude, ChatGPT, Claude Code, Codex, Cursor or VS Code, with or without a terminal: talk through your projects, learn from the guides and the community, start projects and generate sounds.

Maru works with the AI assistant you already use. Once it is connected, you can ask the assistant to start a project, generate sounds, explain how a module works, find ideas in what other creators published or check whether a plug-in build is ready, without leaving the conversation.

The connection uses the Model Context Protocol (MCP). MCP is the standard way assistants such as Claude, ChatGPT and coding agents use outside tools: you add Maru once, and the assistant can then call Maru on your behalf, with the access you approve.

Pick the path that matches how you work. You do not need to write code for either one.

Without a terminal: Claude, ChatGPT or Claude Desktop

Add Maru as a custom connector in the app's settings, using this URL:

text
https://maruaudio.com/mcp
  1. On a computer, open the app's connector settings. In Claude and Claude Desktop: Settings → Connectors → Add custom connector. In ChatGPT, turn on developer mode under Settings → Apps & Connectors → Advanced settings, then choose Create.
  2. Paste the URL above and save.
  3. The first time the assistant uses Maru, it asks you to sign in and approve access in your browser.

That is all. Your Studio projects, generated sounds and plug-in builds are now available to the assistant. Connect your agent has the exact steps for each app, a personal-token option and troubleshooting.

With a coding agent: Claude Code, Codex, Cursor or VS Code

Coding agents can also work on projects on your computer. They need Node.js installed once; download the LTS installer for your system and run it. After that, let the agent do the setup. Paste this into your coding agent:

text
Set up Maru for me. Run `npx maruaudio skill install` and
`npx maruaudio mcp install --both`, then run `npx maruaudio doctor` and tell
me in plain language what is ready and what is missing.

The agent installs the Maru skill (instructions that teach it how Maru projects work) and registers two servers: the hosted connector for your Maru account and a local server for projects in your folder. Restart or reload the agent afterwards. When it first uses the hosted connector it asks you to sign in to Maru in your browser; in Claude Code you can also run /mcp, select maru and choose Authenticate.

If you prefer to run the commands yourself:

sh
npx maruaudio skill install
npx maruaudio mcp install --both
npx maruaudio doctor

Use Maru from your agent explains each piece, the supported agents and where each configuration is written.

What you can ask now

Generate five dusty vinyl kick drums and a matching snare.

Start a sampler bank project from the sounds I just generated.

Find community delays with a tape character and show me how their feedback path is written.

What have I been working on in Studio? Summarize where each project stands.

Why did my last VST3 build fail, and what should I change?

Workflows collects more prompts by task. To start building on your computer, continue with the Quickstart.

How it fits into Maru

Your Studio projects, your plug-in builds and your Generated Sounds live in your Maru account. The hosted connector lets an assistant read and act on them as you, with the access you approve. It also reads the same guides the Maru agents use, so answers about the Engine, Symbolic or the SDK come from the documentation rather than from memory.

The hosted connector reads your project source but does not change it, and it does not build. Editing, checking and building happen on your computer, through the local server and the CLI. A typical coding-agent session uses both: the hosted connector to find the project and read the guide, the local server to change, check and build the code in your folder. Local project files and hosted Studio projects are stored separately; connecting your account does not synchronize a local folder with Studio. A local module can be linked to a Studio project with maru module deploy.

What it can do, and what you approve

  • Know your work. List your Studio projects, then search and read their source to explain, review or plan a change.
  • Answer from the guides. Read the Maru guides for the SDK, the CLI, module authoring and Studio.
  • Learn from the community. Search modules other creators published and read their open source.
  • Track your plug-ins. See which builds finished or failed and why, which versions you released, and which formats install.
  • Start something new. Create a Studio project from a template, or an empty MIDI sound bank.
  • Generate sounds. Render one-shots, loops and 16-pad drum kits from a prompt, saved to your Generated Sounds library. See Generate audio.

The assistant acts as you, with only the access you approve: reading (mcp:read), creating projects and sound banks (mcp:write), and generating audio with your credits (mcp:generate, asked for separately the first time it generates). It cannot publish, read your email or change your account. Tools lists every tool and the permission it needs; the same list is published as JSON at /mcp/tools.json.

Keep it up to date

sh
npx maruaudio mcp status
npx maruaudio mcp remove

status shows which agents have the Maru servers configured, and remove takes them out again. Revoke an assistant's access to your account at any time under Settings → Access tokens, where recent agent activity is listed too.

For the command line and the local server in detail, read Agents and MCP.