The full upstream README, mirrored here for reference. Install config, tool schemas, adoption signals, and an original overview live on the BranderUX MCP listing page.
BranderUX builds agentic applications, in minutes: describe a business and a real agent builds its full application, published as a website and an MCP app, with every screen generated live, per customer, per question. This server is how any MCP client (Claude, ChatGPT, Cursor, VS Code) builds and controls those applications.
This is the official BranderUX MCP server. It gives an AI agent real control over BranderUX projects, brand, custom elements, screens, API keys, plus the verified reference docs it needs to integrate the SDK correctly.
Connect: https://mcp.branderux.com/mcp · Docs: https://branderux.com/mcp
No API keys: the first tool call opens your browser for a one-click BranderUX sign-in (OAuth 2.1 + PKCE, scoped and revocable). Access is currently limited to beta accounts, everyone else lands on the waiting list, and you'll get an email the moment your access opens. While you wait: try a live BranderUX-built app at nova.branderux.app.
Knowledge (no scopes needed, signing in is still required to reach the server):
get_started · read_doc · search_docs · get_integration_snippet
Projects (projects:*): whoami · list_projects · get_project ·
create_project · update_brand_settings · update_project_settings · delete_project
Screens (projects:write): list_screens · get_screen · put_screen ·
delete_screen, reads fetch the project aggregate; writes are atomic per-screen
(row-locked server merge: parallel saves of different screens are safe).
Custom elements (elements:*): list_elements · get_element · list_element_versions ·
create_element · publish_element_version · preview_element · delete_element, your agent writes the
TSX; the server pre-flight validates it (compile + sandbox import allowlist + export
contract) before publishing. In clients that support MCP Apps, create_element,
publish_element_version and preview_element render the element LIVE in the panel,
demo props applied, clicks showing the exact query they would send.
API keys (keys:manage): create_api_key · list_api_keys · set_key_origins ·
revoke_api_key
Hosted agent: upsert_agent_config · get_agent_config, give the project
its own agent: persona, policies (language and clock, login, access, handoff, write consent),
answer quality and budgets.
Business data: define_entity · list_entities · seed_records ·
update_record · list_entity_records, managed records or live store/API feeds; every entity
becomes a query tool for the agent (add-only writes by default).
Connectors: set_connector_credential · probe_api, vaulted credentials and a
probe so tools are wired against the real API shape.
Skills: upsert_skill · list_skills · delete_skill, SKILL.md behavior packs
that ride every conversation.
Home screen: set_home_screen, a designed first paint with live data bindings,
zero model cost per visit.
Publish: publish_site · get_site, one call and the full application is live:
a website for people, an MCP app for AI agents.
Playground (no project needed): generate_screen, renders a real branded,
interactive screen in the panel with demo data, powered by the same published
@brander/mcp-tools package customers install. Ask for a storefront, analytics
or order-flow screen to see actual BranderUX output before building anything.
Every destructive tool requires an explicit confirm: true.
| Env | Default | Purpose |
|---|---|---|
BRANDER_API_BASE | http://localhost:8080/api/v1 | BranderUX API base |
MCP_RESOURCE_URL | http://localhost:3010 | Public URL of this server (OAuth resource id) |
OAUTH_ISSUER_URL | = BRANDER_API_BASE | Authorization server issuer |
The server is stateless: one MCP server instance per request, bound to the caller's
bearer. Deploy target is Vercel (api/mcp.ts + api/oauth-protected-resource.ts).
MIT