> ## Documentation Index
> Fetch the complete documentation index at: https://docs.upsolve.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# MCP Branding

> Present your workspace's MCP server under your own name, icon and loading animation, with canvas links that open in your app.

When your own agent connects to your workspace's MCP endpoint with an [embed token](https://docs.upsolve.ai/ai-agent-builder/deploy-agents/embed-mcp.md), everything your users and your agent see can carry your brand instead of Upsolve's: the server's name and icon, the tool descriptions and instructions your agent's model reads, error messages, and the chart widget in clients like Claude and ChatGPT. Canvas links open in your app.

Set it in **Workspace settings → MCP Branding**. Only organization owners and admins can see the tab and change it.

> [!NOTE]
>
> Branding is not part of a published version: a save applies to the next request, in every environment. A client that is already connected picks it up within 3 minutes.

## Your workspace's MCP URL

The tab shows your workspace's own endpoint, with a copy button:

```
https://mcp.upsolve.ai/w/<WORKSPACE_ID>
```

It takes [embed tokens](https://docs.upsolve.ai/ai-agent-builder/deploy-agents/embed-mcp.md#tokens) only. See [Use Upsolve in your own agent](https://docs.upsolve.ai/ai-agent-builder/deploy-agents/embed-mcp.md) to connect to it.

## Fields

| Field | What it changes | Rules |
| - | - | - |
| **Display name** | The name the server introduces itself with when a client connects, and the name used wherever Upsolve's would be: tool descriptions, instructions, errors and the chart widget. | Up to 80 characters. Empty: **Analytics**. |
| **Description** | Shown by MCP clients beside the server's name. | Up to 300 characters. |
| **Website URL** | Shown by MCP clients beside the server's name. | `https://` only. |
| **Icon** | The server's icon in MCP clients. | An `https://` URL, or a `data:image/...` URI (PNG, JPEG, GIF, WebP or SVG). An SVG icon follows the [loading animation's rules](#loading-animation). Empty: no icon. |
| **Loading animation** | What the chart widget shows while an answer is being written. | See [Loading animation](#loading-animation). |
| **Canvas link** | Where a canvas link opens for your users. | See [Canvas links](#canvas-links). |

Click **Save** to save every field at once. If a value is refused, a message says which field and why, and nothing is saved.

## Loading animation

Clients that draw MCP Apps (Claude, ChatGPT) show the chart widget in the conversation. While the analytics agent works, the widget shows your animation and *"Acme Analytics is working on your answer…"*. Code clients, such as a Pydantic AI agent, never show the widget.

Paste the SVG into the field, or click **Choose file**. **Remove** clears it. With no animation, the widget shows a plain grey spinner.

The animation must be:

* **An SVG file of at most 100 KB**, whose root `<svg>` element has a `viewBox`.
* **Self-contained.** Animate it with CSS (`<style>`, `@keyframes`) or SMIL (`<animate>`, `<animateTransform>`).

These are refused:

* `<script>`, `<foreignObject>`, `<image>`, `<a>` and any other non-SVG element
* event handlers such as `onload` or `onclick`
* links to anything outside the file: an `href` other than `#id`, a `url()` other than `url(#id)`, `@import`, and `data:` URLs
* character references other than `&amp;` `&lt;` `&gt;` `&quot;` `&apos;`
* `<!DOCTYPE>`, `<![CDATA[` and processing instructions

The widget draws the animation as an image, so it can never run code, even in a viewer that ignores these rules.

## Canvas links

When the agent builds or changes a canvas for one of your users, it gives them a link to it. For your users, the link opens your app at the **Canvas link** you set:

```
https://app.acme.com/analytics?canvasId={canvasId}
```

| Placeholder | Replaced with |
| - | - |
| `{canvasId}` | The canvas's ID. Required. |
| `{workspaceId}` | The workspace's ID. |
| `{projectId}` | The project's ID. |

* The link must start with `https://`, and other placeholders are refused.
* Each value is URL-encoded into the link.
* The link carries no credentials, so it's safe to share: only someone who can sign in to your app as that user sees the canvas.
* **With no canvas link,** your users get no link: the agent says the canvas is ready, with nothing to click.
* **Members of your Upsolve organization** keep links to Upsolve, as before.

Your page then has to open the canvas in your [workspace embed](https://docs.upsolve.ai/embedding.md): see [Open a canvas from a link](https://docs.upsolve.ai/embedding.md#open-a-canvas-from-a-link).

## Preview

The right side of the tab shows what your users see, with the values you've typed so far: the server's name and icon, a sample question, and a **Loading** or **Answered** state. **Answered** shows a sample chart and a canvas link built from your template.

## What stays the same

* **Tool names.** They're the same as on Upsolve's own servers, because MCP clients store a user's approvals by tool name.
* **The server's programmatic name**, `analytics`, which clients use as an identifier. Your display name is the server's title.
* **Upsolve's own MCP servers.** `https://mcp.upsolve.ai/mcp` and `/user` are not branded.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.