Connect an AI client with MCP.

Kaitoi Studio is a Model Context Protocol (MCP) server. Connect it to the AI client you already use and that client can search the node catalogue, run a node or a whole graph, build and save projects, upload files and fetch the results. It works in your own Kaitoi account, with your projects, files and credits.

The agent builds graphs in its own working copy. Ask it to save the work as a project and the project opens in Kaitoi like any other. To change an existing project, ask the agent to load it first. A quick single-node run returns its result but does not create a project.

Every client uses one server URL:

https://mcp.studio.kaitoi.io

It is also shown in Kaitoi under Settings → MCP → MCP Access, with a copy button.

#Before you start

MCP access is included in the Indie plan and above. If your account does not have it, the MCP settings page says it is available in Indie, and the server refuses the connection.

There are two ways for a client to sign in:

• OAuth, the recommended way. You add the server URL to your client, the client opens a Kaitoi sign-in page in your browser, and you approve it. You never paste a secret into a config file, and the client renews access on its own, although it may ask you to sign in again from time to time. Use it for every client on this page except Gemini CLI, which is set up with a static token.
• A static token, for scripts, CI and clients that cannot complete a browser sign-in. See Static tokens.

When a client signs in with OAuth you see an Authorize access page that names the client and shows the permissions it asked for. There are two:

Permission	Allows
Read your projects, graphs, nodes, and files	Required. Lets the agent look at your account and the node catalogue.
Create and modify projects, graphs, nodes, files, and runs	Optional. Lets the agent build graphs, upload files and start runs. Untick it for an agent that can look but not change or run anything.

If you are not signed in to Kaitoi in that browser, you are asked to sign in first and then brought back to the same page. Once you click Authorize, the client appears under Settings → MCP → MCP Access → Connected Apps.

Kaitoi can also write the setup for you. Open Settings → MCP → MCP Access → MCP Setup, pick your client, and copy the command or configuration it generates. It covers Claude, Claude Code, Cursor, VS Code, Gemini CLI and Windsurf. For Gemini CLI, use the command in the Gemini CLI section instead, which keeps the token out of your project folder. For ChatGPT, Codex and Zed, follow this page.

#Claude

This covers Claude on the web, Claude Desktop and the Claude mobile apps. Claude connects to Kaitoi from Anthropic's servers, so a connector you add once works in all of them.

1. In Claude, open Customize → Connectors.
2. Click +, then Add custom connector.
3. Name it Kaitoi and paste the server URL. If Claude shows more settings, keep the suggested sign-in options. Click Add.
4. Click Connect, then approve the Kaitoi sign-in in the browser.

On the Free plan Claude allows one custom connector. On Team and Enterprise plans an organization owner adds the connector under Organization settings → Connectors, and members then find it under Customize → Connectors and click Connect.

Use OAuth with Claude. Some organizations have Claude's request header option, which is in beta. There, a static token also works: choose no sign-in and add an Authorization header with the value Bearer <token>. Everyone who uses that connector then shares the token's access.

#ChatGPT

ChatGPT connects to custom MCP servers on the web, not in the mobile apps. Which ChatGPT plans can add one, and whether a workspace admin has to allow it first, is set by OpenAI. Check OpenAI's article on developer mode and MCP apps for the current rules.

OpenAI is partway through renaming apps to plugins, so the labels you see may differ from these:

1. In ChatGPT on the web, open Plugins, click the + button, and choose Create custom MCP server.
2. Name it Kaitoi, add a short description such as "Build and run Kaitoi graphs", and enter the server URL under Connection.
3. Set Authentication to OAuth, accept the warning, and click Create as a plugin. Approve the Kaitoi sign-in when it opens.
4. Review the tools ChatGPT found and install the plugin.
5. Start a new chat, type @ and pick Kaitoi.

If your account still shows the older apps layout, follow OpenAI's article linked above instead: it turns on Developer mode under Settings → Apps → Advanced settings and creates the app from there.

ChatGPT does not accept a static token. Use OAuth.

#Claude Code

Add the server from a terminal:

claude mcp add --transport http --scope user kaitoi https://mcp.studio.kaitoi.io

--scope user makes Kaitoi available in every project. Leave it out to add it to the current project only.

Then start Claude Code, run /mcp, select kaitoi and follow the browser sign-in.

#Codex

The Codex CLI, the Codex IDE extension and the Codex app share one configuration on the same computer, so adding Kaitoi in one adds it in all three. From a terminal:

codex mcp add kaitoi --url https://mcp.studio.kaitoi.io

Codex opens the Kaitoi sign-in in your browser straight away. If you close it, or need to sign in again later, run:

codex mcp login kaitoi

You can do the same without a terminal. In the Codex app, open Settings → MCP servers. In the IDE extension, open its gear menu and choose MCP servers. Then click Add server, name it kaitoi, choose Streamable HTTP, paste the server URL and save. Restart the app, or choose Restart extension in the IDE, then click Authenticate.

The configuration lands in ~/.codex/config.toml:

[mcp_servers.kaitoi]
url = "https://mcp.studio.kaitoi.io"

#Cursor

Add Kaitoi to ~/.cursor/mcp.json to use it in every project, or to .cursor/mcp.json inside one project:

{
  "mcpServers": {
    "kaitoi": {
      "url": "https://mcp.studio.kaitoi.io"
    }
  }
}

Save the file. Cursor lists Kaitoi in its MCP settings and asks you to sign in. Complete the sign-in in the browser.

#VS Code and GitHub Copilot

Open the Command Palette and run MCP: Add Server. Choose HTTP, paste the server URL and name it kaitoi.

To edit the configuration by hand instead, run MCP: Open User Configuration and add:

{
  "servers": {
    "kaitoi": {
      "type": "http",
      "url": "https://mcp.studio.kaitoi.io"
    }
  }
}

VS Code asks you to sign in the first time the server starts. Complete the sign-in in the browser. If your organization manages Copilot, an admin may need to allow MCP servers first.

#Windsurf

Windsurf is now called Devin Desktop. Add Kaitoi to ~/.config/devin/mcp_config.json, or %APPDATA%\devin\mcp_config.json on Windows:

{
  "mcpServers": {
    "kaitoi": {
      "url": "https://mcp.studio.kaitoi.io"
    }
  }
}

Save it. Devin asks you to sign in the first time it uses Kaitoi. If you have the Devin CLI, you can also sign in from a terminal with devin mcp login kaitoi. In the legacy Cascade agent, you can open the same file from the Actions menu at the top of the Cascade panel.

#Gemini CLI

Set up Gemini CLI with a static token. Create one first, as described in Static tokens, and keep it in an environment variable:

export KAITOI_MCP_TOKEN="your token here"
gemini mcp add --transport http --scope user \
  --header "Authorization: Bearer $KAITOI_MCP_TOKEN" \
  kaitoi https://mcp.studio.kaitoi.io

Gemini CLI writes the token into its settings file. --scope user puts it in ~/.gemini/settings.json rather than in a .gemini folder inside the current project, where it could end up committed to a repository.

Run gemini mcp list to confirm Kaitoi is connected.

#Zed

Open Settings → AI → MCP Servers, click Add Server, then Add Remote Server, and paste the server URL. In settings.json the result looks like this:

{
  "context_servers": {
    "kaitoi": {
      "url": "https://mcp.studio.kaitoi.io"
    }
  }
}

Zed asks you to sign in when the server has no Authorization header configured. Complete the sign-in in the browser.

#Other clients

A client needs to support remote MCP servers over Streamable HTTP, and an OAuth sign-in that works with Kaitoi. Add a remote server, name it Kaitoi, give it the server URL, and complete the browser sign-in. Kaitoi supports dynamic client registration, so clients that register themselves automatically need no client ID or secret.

If the sign-in does not complete, and the client can send an Authorization header, use a static token instead.

#Static tokens

A static token is for CI, scripts, headless machines and clients that cannot complete a browser sign-in. Prefer OAuth everywhere else: a static token does not renew itself, and anyone who has it can use it until it expires a year after you create it, or until you revoke it.

To create one:

1. In Kaitoi, open Settings → MCP → MCP Access.
2. Click New MCP Token, name it after the client that will use it, and choose its permissions.
3. Click Create and copy the token. It is shown once.

Create one token per client, so you can revoke any of them without breaking the others. Tokens only work with the MCP server. They are not Kaitoi Core API keys.

Clients send the token as a header:

Authorization: Bearer <token>

The MCP Setup dialog can generate a full command or configuration with a fresh token for most clients, under Manual configuration with a static token. For Gemini CLI, use the command in the Gemini CLI section rather than the generated one. To write it yourself, keep the token in an environment variable where the client supports one.

Claude Code:

claude mcp add --transport http --scope user kaitoi https://mcp.studio.kaitoi.io \
  --header "Authorization: Bearer $KAITOI_MCP_TOKEN"

Codex, which reads the token from the named variable each time it connects:

codex mcp add kaitoi --url https://mcp.studio.kaitoi.io \
  --bearer-token-env-var KAITOI_MCP_TOKEN

Cursor:

{
  "mcpServers": {
    "kaitoi": {
      "url": "https://mcp.studio.kaitoi.io",
      "headers": {
        "Authorization": "Bearer ${env:KAITOI_MCP_TOKEN}"
      }
    }
  }
}

#What your agent can do

Once connected, the agent has tools to:

• search the node catalogue and read a node's inputs and outputs
• run a single node, or build a graph by adding and connecting nodes, then validate and run it
• list, load and save your projects
• import a file from a URL into your Kaitoi storage and use it as an input
• fetch the images, video, audio, 3D models and text a run produced
• on plans with raw code editing, write new node types

Runs started this way are billed to your Kaitoi account, the same as runs started on the canvas. Kaitoi does not stop paid runs for your review. Some quick single-node runs return a cost preview that the agent has to confirm, but it can confirm without asking you, and other runs, including graph runs, start without one. Many clients can ask you to approve each tool call; turn that on if you want to see runs before they start. For an agent that can explore but never spend your credits, grant it only the read permission.

Ask for Kaitoi by name the first few times, for example "use Kaitoi to find an image upscaler". Some clients otherwise reach for a web search.

#Disconnecting

• OAuth clients: open Settings → MCP → MCP Access, find the client under Connected Apps, and click Disconnect. To change a client's permissions, disconnect it and connect it again: you choose them afresh on the Authorize access page.
• Static tokens: find the token under Access Tokens on the same page. You can change its permissions there, or revoke it.

Remove Kaitoi from the client's own settings as well, or it will keep trying to reconnect.

#Troubleshooting

The MCP settings page says it is available in Indie. Your account does not have MCP access. Check your plan: MCP access is included from Indie up.

The client reports an invalid token or keeps asking to reconnect. With OAuth, clear the client's saved Kaitoi sign-in and connect again. In Claude Code, run /mcp, select kaitoi and clear its authentication. In Codex, run codex mcp logout kaitoi and then codex mcp login kaitoi. Elsewhere, remove the server and add it again. With a static token, check the token and any environment variable the client reads it from. If the token has expired or been revoked, create a new one under Access Tokens and update the client.

The agent can read but cannot run anything. Check that the connection has the write permission. For an OAuth client, disconnect it under Connected Apps, connect it again and leave the second permission ticked. For a static token, add the permission under Access Tokens. If write is already granted, check the client's own tool settings and the error the run returned.

The agent says Kaitoi is unavailable, or searches the web instead. Check the connector is switched on for the conversation, then mention Kaitoi by name. In ChatGPT, pick it with @.

#Current limitations

• ChatGPT connects with OAuth only. So does Claude, apart from organizations with its request header beta.
• Gemini CLI is set up with a static token rather than a browser sign-in.
• Static tokens do not renew. Each one expires a year after you create it, so replace it before then, and revoke any token you no longer use.