> For the complete documentation index, see [llms.txt](https://docs.arcade.software/kb/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.arcade.software/kb/ai-video/mcp.md).

# MCP

### What it does

The Arcade MCP server connects AI tools — Claude, ChatGPT, Gemini, Cursor, and others — to your Arcade workspace, so you can create Brand Kits and generate AI videos through natural conversation instead of the composer UI. Videos are generated end-to-end and typically ready in 1–4 minutes.

### Prerequisites

An Arcade account with a workspace ([sign up at app.arcade.software](https://app.arcade.software)).

### How do I connect my AI tool?

Every client connects to the same server URL, `https://mcp.arcade.software/mcp`, then signs in with your Arcade account to authorize access. See [Supported Clients](https://docs.arcade.software/kb/ai-video/mcp/supported-clients.md) for exact steps per tool — Claude Web, Claude Desktop, Claude Code, ChatGPT, Gemini, Cursor, Codex App, Lovable, Runlayer, the Cloudflare MCP Portal, and any other MCP client that supports Streamable HTTP and OAuth.

### Do I need a paid plan?

Arcade works on all plans (Free, Pro, and Enterprise) — creating a Brand Kit or generating a video through MCP isn't gated by plan.

### How do permissions work?

When connecting Arcade, you'll be prompted to grant permissions. You can choose to:

* Allow once
* Allow always
* Require approval

Permissions are grouped into read actions (browsing your Brand Kits and demos) and write actions (creating a Brand Kit, generating a video). Delete actions are not currently supported.

### What can I ask it to do?

See [Use Cases](https://docs.arcade.software/kb/ai-video/mcp/use-cases.md) for real examples — launch videos from recent releases, five channel-specific variants from one brief, personalized outreach videos from call notes, and more.

### Can I create, edit, and get the finished file all through MCP?

Yes, the whole loop stays in your AI tool:

1. Create a Brand Kit (`create_brand_kit`) if you don't have one ready — this is optional going in: `create_video` uses your named kit, your workspace's default, or your only kit automatically if you skip it, and tells you what's missing if none of those exist
2. Generate a video (`create_video`)
3. Refine it with follow-up instructions in the same chat (`edit_video`)
4. Call `preview_video` — it returns both the exported MP4's own URL and a link to open the video in the Arcade editor, so you don't have to leave your AI tool to get the finished file

Check your credit balance before or during this with `get_credits`. See [Available Tools](https://docs.arcade.software/kb/ai-video/mcp/available-tools.md) for the complete tool list.

### What happens if I run out of AI credits?

Video generation uses AI credits — see [Plans, Credits & Pricing](https://docs.arcade.software/kb/ai-video/plans-credits-and-pricing.md). If you run out, requests fail until you upgrade or your balance refreshes. Call `get_credits` to check your balance, or what a specific generation would cost, before you start.

***

## Security

### Does Arcade train on my data?

No. Arcade does not train on your data.

### What happens when I connect Arcade to third-party tools like Claude?

When you connect Arcade to an external AI tool, data shared through that workflow becomes subject to the third party's policies. This is especially important when using free AI plans or working with sensitive information.

### Are free AI tools safe for sensitive data?

Free AI tools may store inputs, use data to improve their systems, or have different retention policies. For sensitive use cases, consider using paid or enterprise AI plans and review the provider's data policies.

### How does data access actually work?

Claude does not directly browse your Arcade dashboard. Instead, Claude calls specific MCP tools, and those tools return scoped data based on the request — this creates more controlled access than open dashboard access would.

### Can I control what the MCP can do?

Yes — when you connect, you choose `Allow once`, `Allow always`, or `Require approval` for each group of permissions (see How do permissions work? above), rather than granting blanket access. To change your answer later, manage or re-authorize the connection from your AI tool's own integrations/connectors settings — Arcade doesn't have a separate permissions screen of its own.

### What should enterprise teams consider?

For sensitive use cases:

* Use enterprise AI plans
* Review third-party security policies
* Limit permissions where needed
* Follow your internal governance standards

***

## Limitations

* Cannot create or edit interactive demos — that's a separate product with its own MCP tools, including reading an existing demo's performance (views, completions, drop-off, trending demos). See Interactive Demo's [MCP](https://docs.arcade.software/kb/interactive-demo/leverage/mcp/) section for those.
* Editing through MCP (`edit_video`) only works on a video that MCP itself created — `create_video` or a previous `edit_video` — anywhere in your workspace, not just earlier in the same chat; it can't touch a video made directly in the Arcade editor
* `create_video`'s length option tops out at 120 seconds — a separate, lower ceiling than the composer's plan-based caps (up to 5 minutes on Pro, 10 on Enterprise); for a longer video, generate it in the composer instead
* The inline video player (`preview_video`) only works on AI tools that can render rich widgets today — Claude and ChatGPT. Other clients get a link to open the video instead
* Video creation depends on AI credits
* Output quality may vary based on prompt quality
* Behavior may vary depending on the AI client being used

## Troubleshooting

| Issue                                      | Solution                                                                                                                |
| ------------------------------------------ | ----------------------------------------------------------------------------------------------------------------------- |
| "Brand kit not found"                      | The brand kit ID may be incorrect, or the kit isn't ready yet. Use `list_brand_kits` to check status before generating. |
| Video generation times out or stalls       | Generation can take a few minutes. Ask the assistant to check again — it picks up right where it left off.              |
| "This operation requires generation scope" | Reconnect the MCP server — your authorization may have expired or didn't grant write access.                            |
| Brand kit stuck on "generating"            | Typically takes 2–4 minutes. If it's been longer than 10 minutes, try creating a new one.                               |
| Source addition fails                      | Verify the URL is publicly accessible, not behind a login — the scraper needs to reach the page.                        |
| "Generation failed"                        | The generation hit an error and can't be recovered — start a new one.                                                   |
| Connection drops mid-generation            | Your video keeps generating in the background. Ask the assistant to check its status again rather than starting over.   |


---

# 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://docs.arcade.software/kb/ai-video/mcp.md?ask=<question>&goal=<user_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.
