Brand MCP server

The Pulumi Brand MCP Server is a remote Model Context Protocol server that exposes the Pulumi brand guidelines as structured resources, tools, and prompts. It’s designed to be consumed by AI agents and LLM-powered workflows that create, review, or audit brand content, and it’s deployed at brand.pulumi.com and compatible with any MCP client that supports streamable HTTP.

Configuration

Point any MCP-compatible client at the public URL:

https://brand.pulumi.com/mcp

In the snippets below, we’ve named the connection pulumi-brand, but you can change that if you like.

Claude

You can use the MCP server with Claude Code, Claude Desktop, and in the browser at Claude.ai.

Claude Code

Add the MCP server with the CLI:

claude mcp add --transport http pulumi-brand https://brand.pulumi.com/mcp

By default, the server is scoped to the current project. Add --scope user to make it available across all of your projects:

claude mcp add --transport http --scope user pulumi-brand https://brand.pulumi.com/mcp

To confirm the connection succeeded, run claude mcp list. To inspect the server’s tools, run /mcp in an active Claude Code session.

Claude Desktop

You can configure Claude Desktop with the MCP server in one of two ways.

With a custom connector

In Claude Desktop, open Customize → Connectors.
Click Add → Add custom connector.
Enter a name (e.g., “Pulumi Brand”) and paste the URL: https://brand.pulumi.com/mcp.
Save the connector and begin using the MCP server.

With a local MCP server

This approach uses the mcp-remote connector.

In Claude Desktop, open Developer → Local MCP servers.
Click Edit config and open the file in your text editor of choice.

Add a pulumi-brand block to the mcpServers list (add that list if it doesn’t exist):

{
    "mcpServers": {
        "pulumi-brand": {
            "command": "npx",
            "args": ["-y", "mcp-remote", "https://brand.pulumi.com/mcp"]
        }
    }
}

Save the file and begin using the MCP server.

Claude.ai

Visit https://claude.ai, sign in, and navigate to Customize → Connectors.
Click Add → Add custom connector.
Enter a name (e.g., “Pulumi Brand”) and paste the URL: https://brand.pulumi.com/mcp.
Save the connector and begin using the MCP server.

Codex

Add the following block to ~/.codex/config.toml for all projects, or to ./.codex/config.toml for the current one:
```
[mcp_servers.pulumi-brand]
url = "https://brand.pulumi.com/mcp"
```
Save the file, restart Codex, and begin using the MCP server.

Cursor

You can use the MCP server with Cursor in several ways.

Cursor Desktop

Open Settings → Cursor Settings → Tools & MCPs.
Under Installed MCP Servers, click Add Custom MCP.

Add the following block to ~/.cursor/mcp.json for all projects, or to ./.cursor/mcp.json for the current one:

{
    "mcpServers": {
        "pulumi-brand": {
            "transport": "http",
            "url": "https://brand.pulumi.com/mcp"
        }
    }
}

Save the file and begin using the MCP server.

Cursor CLI

Add the following block to ~/.cursor/mcp.json for all projects, or to ./.cursor/mcp.json for the current one:

{
    "mcpServers": {
        "pulumi-brand": {
            "transport": "http",
            "url": "https://brand.pulumi.com/mcp"
        }
    }
}

Save the file, restart Cursor, and begin using the MCP server.

OpenCode

Add the following block to ~/.config/opencode/opencode.json for all projects, or to ./opencode.json for the current one:

{
    "$schema": "https://opencode.ai/config.json",
    "mcp": {
        "pulumi-brand": {
            "type": "remote",
            "url": "https://brand.pulumi.com/mcp",
            "enabled": true
        }
    }
}

Save the file, restart OpenCode, and begin using the MCP server.

Other clients

Any MCP client that supports a remote Streamable HTTP server works. Give it the URL https://brand.pulumi.com/mcp directly if it accepts one, or bridge through stdio with npx -y mcp-remote https://brand.pulumi.com/mcp (the same pattern as the Claude Desktop config-file option above).

Resources

Static brand content served via resources/read. Each resource uses a brand:// URI scheme.

URI	Name	Content
`brand://colors`	Brand Colors	Full color system — 8 palettes × 11 steps, semantic roles, utility colors, accessibility notes
`brand://colors/{palette}`	Brand Color Palette	A single palette (e.g., `brand://colors/violet`)
`brand://typography`	Brand Typography	Inter + Monaspace Neon typefaces, font stacks, usage rules, typographic scale
`brand://voice`	Brand Voice & Tone	Voice principles, tone-by-context matrix, preferred/avoided words, inclusive language
`brand://writing-style`	Writing Style	Grammar, punctuation, style rules (Chicago Manual of Style)
`brand://logo`	Logo Guidelines	Lockups, sizing, approved backgrounds, clear space, dos/don’ts
`brand://identity`	Brand Identity	Brand attributes, purpose, values, Phosphor iconography guidelines
`brand://visuals`	Visual Guidelines	Illustration style, accent lines, photography, mascot (Pulumipus)
`brand://guidelines`	Usage Guidelines	Generative AI usage rules, when to bend brand rules

Tools

The tools fall into two groups: brand lookup (colors, typography, search) and asset generation (logos).

`get_brand_overview`

Return the brand orientation — how to use this server and which get_guidelines section to pull for the task at hand. The same text MCP clients receive as server instructions at session start; call this when your client doesn’t surface those, or to re-orient mid-task. It routes you to the rules rather than restating them — use get_guidelines for the full guidance on a topic.

Takes no arguments.

`get_guidelines`

Return the complete text of a brand guideline section — the full guidance, not a summary. Call it before producing or reviewing the matching kind of artifact (for example, voice and writing-style before writing copy). For color, use the color tools below; to search across sections, use search_guidelines.

Parameter	Type	Required	Description
`section`	string	Yes	One of: `voice`, `writing-style`, `typography`, `logo`, `identity`, `visuals`, `guidelines`

`get_color_palette`

Get a specific brand color palette with all 11 steps (50–950) and semantic roles.

Parameter	Type	Required	Description
`palette`	string	Yes	One of: `violet`, `gray`, `green`, `aqua`, `blue`, `red`, `orange`, `yellow`, or `utility` (black, white, Service Black)
`format`	string	No	Color format hint: `hex`, `hsl`, `rgb`, `oklch`. Stored values are hex; other formats available via `brand://colors`.
`response_format`	string	No	`json` (default) for structured data or `markdown` for a human-readable table

`find_nearest_brand_color`

Find the perceptually closest brand color to any CSS color value using OKLab Euclidean color difference.

Parameter	Type	Required	Description
`color`	string	Yes	Any CSS color: `#6366f1`, `rgb(99,102,241)`, `hsl(239,84%,67%)`, `oklch(...)`, or a named color
`mode`	string	No	`light`, `dark`, or `both` (default). Which mode’s palette to match against.

Returns the matched token name, palette, step, mode, and hex value with a replacement recommendation.

`get_color_tokens`

Get semantic color tokens (primary, accent, muted, background) for one or all palettes.

Parameter	Type	Required	Description
`palette`	string	No	A specific palette name. Omit to get tokens for all 8 palettes.

`check_color_accessibility`

Check contrast between two colors using WCAG 3 APCA (Accessible Perceptual Contrast Algorithm). Returns the Lc value, polarity, pass/fail thresholds, and a font-size lookup per weight.

Parameter	Type	Required	Description
`foreground`	string	Yes	Hex value or brand token name (e.g., `violet-700-light`)
`background`	string	Yes	Hex value or brand token name

APCA thresholds:

Lc (absolute)	Use case
≥ 90	Preferred body text
≥ 75	Body text minimum
≥ 60	Large / heading text
≥ 45	Large non-text elements
≥ 30	Non-text / spot elements
< 30	Insufficient — do not use

`get_font_stack`

Get the correct brand font stack and usage rules for a given typographic context.

Parameter	Type	Required	Description
`context`	string	Yes	`heading` (Inter Semibold), `body` (Inter Regular), or `mono` (Monaspace Neon)

`search_guidelines`

Full-text search across all brand guideline content. Returns excerpts with surrounding context.

Parameter	Type	Required	Description
`query`	string	Yes	Search terms or natural language question about the brand

`get_logo`

Return a URL for a Pulumi logo in any approved variant, format, background, and size. The URL is served by the brand site’s logo endpoint at brand.pulumi.com/api/logo (CDN-cached) — surface it directly to the user. Pass inline: true to embed the bytes (SVG markup or a base64 data URI) in the tool response instead, for runtimes that can’t fetch a URL at render time.

Parameter	Type	Required	Description
`usage`	string	No	Preset that sets sensible defaults: `web`, `print`, `slide-deck`, `email-signature`, `favicon`, `social-avatar`, `custom`. Explicit params override.
`lockup`	string	No	`auto` (default), `mark`, `horizontal`, `stacked`, or `icon-circle` / `icon-rounded` / `icon-square`
`treatment`	string	No	`color` (default) or `monotone`. Icons are always color.
`theme`	string	No	`light` or `dark`. Auto-derived from the background when omitted.
`format`	string	No	`svg` (default), `png`, `jpg`, or `pdf`
`background`	string	No	`transparent` (default), `white`, `black`, `service-black`, `violet-50`, `violet-950`
`width` / `height`	number	No	Size in `units`. One alone → logo-only; both → centered on a padded canvas.
`units`	string	No	`px` (default), `cm`, or `in` (with `dpi`, default 300)
`inline`	boolean	No	`true` to embed the asset bytes in the response instead of returning a URL. Default `false`.

Prompts

Pre-built prompt templates that inject brand context for LLM-powered workflows.

Prompts are user-invoked — the agent doesn’t call them automatically. You trigger one, fill in its arguments, and it expands into a ready-made message (pre-loaded with the relevant brand guidelines) that’s sent to the model. How you invoke them depends on your client:

Claude Code — they appear as slash commands. Type / and look for the pulumi-brand entries (e.g., /mcp__pulumi-brand__review_copy).
Claude Desktop / Claude.ai — open the + (connectors) menu in the composer, pick the pulumi-brand server, choose the prompt, and fill in the fields.
Other clients — wherever the client surfaces MCP prompts (a command palette or MCP panel). Under the hood it’s the standard prompts/list → prompts/get flow.

Note that review_image takes no arguments — attach the image to your message first, then invoke it.

`review_copy`

Review text content for compliance with brand voice, tone, and writing style. Returns an overall assessment, a list of specific issues with quotes and suggested rewrites, and a clean rewritten version if issues were found.

Argument	Required	Description
`copy`	Yes	The text content to review
`context`	No	`product`, `marketing` (default), `community`, or `enterprise`

`review_image`

Evaluate an image against brand visual guidelines — color, typography, iconography, illustration, and photography. Attach the image to your message, then invoke the prompt.

Takes no arguments — the prompt reads the attached image directly.

`review_design`

Review a design in Figma for brand adherence — covers colors, typography, logo usage, spacing, and overall alignment. Works in combination with the Figma MCP, which retrieves the design from the URL.

Argument	Required	Description
`figma_url`	Yes	Figma file or node URL (`figma.com/design/…`) to review