Brand Stack MCP Server — Connector for Claude, Cursor, and Codex
The Brand Stack MCP server (Model Context Protocol) connects Claude, Claude Code, Claude Desktop, Cursor, Cowork, and Codex to your Brand Stack workspace via OAuth 2.1. Read brands, products, and personas, and turn screenshots into mockups, from any AI chat.
The Brand Stack MCP server lets any Model Context Protocol-compatible AI client (Claude.ai, Claude Code, Claude Desktop, Cursor, Cowork, Codex, and others) read your Brand Stack workspace and turn screenshots into mockups without leaving the conversation.
Quick reference
| MCP endpoint | https://www.brandstack.dev/api/mcp |
| Transport | Streamable HTTP (MCP 2025-06-18) |
| Auth | OAuth 2.1 with PKCE (connectors); API keys for the CLI and scripts |
| Discovery | https://www.brandstack.dev/.well-known/mcp.json |
| Tool count | 15 (11 read + 4 write) |
Once connected, your AI client can:
- Read every brand profile, product, persona, project, document, and asset in the workspace.
- Export brand profiles as Markdown, CSS variables, Tailwind config, or prompt-ready context.
- Generate new product mockups and lifestyle scenes that consume credits from the connected workspace, just like the in-app studio. A newly connected workspace’s first 10 mockups are free.
- Turn screenshots into mockups, the flow coding agents connect for: a Playwright capture uploaded from the agent’s shell, a pasted URL captured server-side (
capture_screenshot), or a pasted image handed over via a loginless upload link. See Screenshots → mockups. - Browse the asset catalog with filters by source, product, and free-text search.
How it works
Brand Stack is an OAuth 2.1 authorization server with a single MCP endpoint at https://www.brandstack.dev/api/mcp. When you add it to your AI client, the client kicks off a one-time browser flow:
- The client registers itself with Brand Stack via Dynamic Client Registration (RFC 7591).
- Your browser opens to a Brand Stack consent screen. Sign in if you aren’t already.
- Pick a workspace from the dropdown (admin-only).
- The client receives a short-lived access token plus a 30-day refresh token.
After that the client talks to /api/mcp over plain HTTPS with a bearer token. No keys to copy and paste, no secrets in your shell history.
Tokens at a glance
| Token | TTL | What it does |
|---|---|---|
| Access token | 1 hour | Authenticates each MCP request. Hashed at rest with SHA-256. |
| Refresh token | 30 days | Rotates on every use. Reuse triggers automatic revocation of the entire grant. |
You can revoke any client at any time from Settings → Developer → Authorized clients. Revocation kills every active access and refresh token for that grant within seconds.
Scopes
Tokens carry one or both of these scopes:
mcp:read: read brands, products, personas, projects, documents, the image library, and credits.mcp:write: capture and upload screenshots (free), and create mockups and scenes (consumes credits; a new workspace’s first 10 mockups are free).
Both scopes are granted by default during the consent flow.
What’s next
- Setup guide: step-by-step instructions for Claude Code, Claude.ai web, Cursor, Codex, and any other MCP-compatible client.
- Example prompts: copyable prompts to try once you’re connected.
- Tools reference: every tool the connector exposes, with input parameters and example calls.
- Troubleshooting: fixes for common issues with the OAuth flow, scopes, and tool errors.