# Brand context for agents over MCP

> Which brand context APIs have a Model Context Protocol server for Claude, ChatGPT or Cursor, the tools each one has, how they sign in, and what agents get back.

Every vendor on this site runs a Model Context Protocol (MCP) server. An agent in Claude, ChatGPT, Cursor or any MCP client can call it without code. Add the address of the server in the connector settings of the client. Sign in, and then ask in plain words.

| Vendor | Server | Sign-in | Tools for brand context |
| --- | --- | --- | --- |
| Brandfetch | `https://mcp.brandfetch.io/mcp` | OAuth, or an MCP token from the dashboard | `get_brand_context` (voice, positioning), `get_brand_data` (logos, colors, fonts) |
| Context.dev | `https://mcp.context.dev/mcp` | OAuth, or an API key | Not listed in its setup documentation. Smithery lists `get-brand` |
| Firecrawl | `https://mcp.firecrawl.dev/v2/mcp` | An API key, browser sign-in, or none with fewer tools | `firecrawl_scrape`. Firecrawl does not document the branding format through this tool |
| Logo.dev | `https://mcp.logo.dev/mcp` | OAuth, no key in the client | `brand`, plus `logo` and `brandmark` |
| Rainbrand | `https://mcp.rainbrand.com/mcp` | Sign in to Rainbrand from the agent | `create_brief` and `get_brief` for a task, `get_brand` for the whole brand |

## What to ask for

The tool returns the same data as the API, so the advice on the [Fields](https://brandcontextapi.com/fields) pages applies. The difference is that the agent decides how to use the data. Tell the agent how.

For a company that you do not work with, through Brandfetch:

```text
Get the brand context for linear.app. Use its voice and the tones to avoid
when you draft the email below, and do not repeat any prices or numbers from it.
```

For a brand you or your client keep in Rainbrand:

```text
Request a Brief for Linear for this task: a hero section announcing Loops,
headline, subhead and call to action. Follow its guidance and claim limits,
and use only the assets it returns.
```

Rainbrand supplies the guidance and assets for the task. The agent that you work in still writes and makes the work. An agency can keep each client as its own brand and name the brand in each request.

## Watch for

- Response size: Context.dev limits MCP responses to 80,000 characters. A full Styleguide is large.
- Billing: agents call tools more often than people expect. Brandfetch, Context.dev, Logo.dev and Rainbrand bill MCP calls like API calls. The Firecrawl documentation does not say.
- Waiting: a Rainbrand Brief is not instant, and ours took 42 seconds. The Rainbrand documentation says to follow progress with `get_brief`. A new request can create a duplicate.

The [MCP quickstart](https://docs.rainbrand.com/mcp/getting-started?utm_source=brandcontextapi.com&utm_medium=satellite&utm_campaign=recipe-mcp) shows how to connect Rainbrand to Claude, ChatGPT, Cursor and other agents.

---

Source: https://brandcontextapi.com/recipes/agents-over-mcp
Last updated: 2026-11-12
Published by The Brand Object. The Brand Object is run by Rainbrand, which makes two of the endpoints described here. Their pages follow the same rules as the others.