> For the complete documentation index, see [llms.txt](https://help.intangible.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://help.intangible.ai/intangible-mcp-agents/intangible-mcp-agents.md).
# Agents & MCP
Connect Claude, ChatGPT, Codex, Grok, Muse, Gemini, Cursor, Perplexity or an open model to Intangible over MCP. Setup steps, then your first scene.
Connect the AI agent you already use, such as Claude, ChatGPT, Grok or Muse, to Intangible over MCP, and direct Intangible from it. Your agent can build and edit 3D scenes, and direct the agents and actions within Intangible.
Your agent builds 3D scenes in your Studio tab from a plain-language description. It works through Intangible's Model Context Protocol (MCP) server. Your agent turns your description into code that builds the scene. The scene fills in while you watch.
{% hint style="info" %}
**MCP server URL:** `https://mcp.intangible.ai/mcp`
{% endhint %}
Intangible's MCP server opens Build, Compose and Visualize to agents. See [What your agent can do](/intangible-mcp-agents/what-your-agent-can-do.md).
## 3D scenes as code
Intangible treats a 3D design as code. Your agent writes code, including math and procedural layouts, to build the scene. Intangible's MCP server runs that code safely in a sandbox. One piece of code can lay out a whole room or a full grid of desks, so scenes build quickly and at scale.
You still direct in plain language. The agent writes the code. See [Scene scripts](/intangible-mcp-agents/what-your-agent-can-do.md#scene-scripts-3d-scenes-as-code).
## Which agents work with Intangible
Each company page lists the apps and models that work, with setup steps:
| Company | Apps and models |
| ---------------------------------------------------------- | -------------------------------------------------------- |
| [Anthropic](/intangible-mcp-agents/anthropic-claude.md) | Claude (claude.ai, Desktop, mobile) and Claude Code |
| [OpenAI](/intangible-mcp-agents/openai.md) | ChatGPT and Codex |
| [xAI](/intangible-mcp-agents/connect-grok.md) | grok.com, Grok Bot and Grok Build |
| [Meta](/intangible-mcp-agents/connect-muse.md) | The Muse app and Muse Code |
| [Google](/intangible-mcp-agents/google-gemini.md) | The Gemini app (personal Google accounts) and Gemini CLI |
| [NVIDIA](/intangible-mcp-agents/nvidia-nemotron.md) | Nemotron models, run through Grok Build |
| [Cursor](/intangible-mcp-agents/connect-cursor.md) | Cursor, in Agent mode |
| [Perplexity](/intangible-mcp-agents/connect-perplexity.md) | Perplexity, as a Computer task |
Any other MCP client connects if it supports remote servers over streamable HTTP with OAuth sign-in. See [Other MCP clients](/intangible-mcp-agents/connect-other-agents.md).
## Before you start
* An Intangible account. Sign up at [studio.intangible.ai](https://studio.intangible.ai/?utm_source=help.intangible.ai\&utm_medium=docs\&utm_campaign=inline-link\&utm_content=intangible-mcp-agents) – see [Sign up and set up](/get-started/sign-up-and-set-up.md).
* The agent app you want to use, installed and signed in.
* A project open in Studio, in a single browser tab.
## For AI agents setting up Intangible
These pages are written so an AI agent can follow them for its user. Read any page as Markdown by adding `.md` to its URL. The index of every page is at `https://help.intangible.ai/llms.txt`, and the full text of the site is at `https://help.intangible.ai/llms-full.txt`. To search these docs as tool calls, connect the docs MCP server at `https://help.intangible.ai/~gitbook/mcp`: it has two tools, `searchDocumentation` and `getPage`. It answers questions about these docs; it doesn't control Studio. If you're an agent:
1. Check whether Intangible's tools are already available to you, for example `get_or_set_studio_target` and `load_skill`. If they are, call `load_skill('intangible-agent-orientation')` and follow it.
2. If they aren't, find the user's app in the table above and open its page. Run the steps you can run, such as editing a settings file or a terminal command.
3. Hand the user the steps only they can do: clicking **Approve access** on the Intangible page, consent dialogs, and risk checkboxes. Name the exact button.
4. Use these values. The server URL is `https://mcp.intangible.ai/mcp`. The transport is streamable HTTP. Sign-in is OAuth with dynamic client registration, so there's no API key, client ID or secret.
5. Before you build, make sure exactly one Studio tab is open. Confirm the project with `get_or_set_studio_target`.
6. Build with code. Call `load_skill('intangible-scene-scripting')` and follow it.
7. Don't delete scene content unless the user asks you to.
{% hint style="warning" %}
**Keep one Studio tab open at a time.** The agent builds in the Studio tab open in your browser. With two or more open, it can build in the wrong project, or stop because it can't tell which one you mean. Close every other Studio tab, including the Studio home page, before you send your first prompt.
{% endhint %}
## How connecting works
Every agent connects the same way, whatever the menu is called. Your company's page has the exact clicks or commands:
1. **Add the server.** Give your agent the URL `https://mcp.intangible.ai/mcp` as a remote MCP server. Apps call it a connector, a plugin, a custom app or an MCP server.
2. **Sign in.** An Intangible page opens that names the agent asking, for example *"Claude is requesting access to your Intangible account"*. Click **Approve access**. There is no API key to copy.
3. **Open one Studio tab.** The agent builds in the Studio tab open in your browser. Keep exactly one project tab open. With two, some agents can't tell which scene to edit.
4. **Describe your scene.** The agent loads Intangible's orientation first, then plans, builds, and checks in with you.
## Describe your first scene
Open a project in Studio, put your agent's chat next to it, and describe the space. Start small: one room or one zone.
Say three things:
* **What the space is.** "A small living room", "a product plinth on a stage", "a hotel lobby".
* **What's in it.** Name the main objects and how many.
* **How they relate.** Use plain spatial words: in front of, behind, left of, on, facing.
For example:
> Use Intangible. My Studio tab is open on my project. Build a small living room: a sofa facing a TV on a low stand, a coffee table between them, and a floor lamp in the corner behind the sofa.
What you'll see:
1. The agent's first reply orients you: chat on one side, Studio on the other, and the order Build, then Compose, then Visualize.
2. It searches the asset library and places the objects. A five-object room takes one to four minutes.
3. The objects appear in your Studio tab as grey models, and the agent lists what it placed and where.
4. It may name a stand-in, such as a media console for a TV stand, or add walls and a floor. Tell it to swap or remove anything you don't want.
Then keep directing in plain language, one change per message:
* **Build out:** "Add a rug under the coffee table." "Put two armchairs facing the sofa."
* **Frame a shot:** "Frame a wide shot of the room from behind the TV, 35mm."
* **Render:** "Render that shot, photoreal, warm evening light."
**Let your agent build the scene.**
Create a free Intangible account
## What to know before a big build
* **Pick the largest model your plan offers.** Bigger scenes reward bigger models. See [Choosing a model](/intangible-mcp-agents/choosing-a-model.md).
* **Approvals.** Some agents ask before every step. Claude and Cursor let you allow the Intangible tools once; Perplexity asks per step.
* **Sign-ins expire.** If a connection that worked yesterday stops, sign in again from your agent. Codex says its access token could not be refreshed when this happens.
* **Teams.** If you belong to more than one Intangible team, the agent asks which team to create a project in.
* **Timing.** A large scene takes minutes. The viewport stays empty for the first couple of minutes while the agent searches the asset library.
* **Objects land close.** Make the last nudges by hand in Build mode, such as which way a chair faces. That's faster than another prompt.
## Start from a loose idea
You don't need a finished brief. Open with a loose idea, such as "a launch film for our new electric car". The agent asks one round of questions about the purpose, the rough scale and what matters most, then builds. See [What your agent can do](/intangible-mcp-agents/what-your-agent-can-do.md#start-from-a-loose-idea).
## Related
* [Choosing a model](/intangible-mcp-agents/choosing-a-model.md)
* [What your agent can do](/intangible-mcp-agents/what-your-agent-can-do.md)
* [Troubleshooting and FAQ](/intangible-mcp-agents/troubleshooting-and-faq.md)
*Last updated 23 October 2026. Steps and capabilities may change.*
---
# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.
## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation by asking a question.
Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:
```
GET https://help.intangible.ai/intangible-mcp-agents/intangible-mcp-agents.md?ask=&goal=
```
`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.
Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.