@craveup/mcp package. It lets AI agents use Crave-specific tools instead of guessing API shapes from memory.
Use it when you want an agent to:
- inspect or import menus
- connect a storefront project to a Crave location
- check storefront readiness before launch
- configure location ordering settings
- work through go-live onboarding steps
- fetch Crave implementation guidance while building a storefront
Setup
The MCP server runs over stdio. Use the public npm package for normal usage outside this monorepo, or point your client at the local repo build when testing unreleased tools. You do not need a clone ofcraveup-turborepo to use the MCP server — npx -y @craveup/mcp fetches everything it needs.
Every configuration below sets CRAVEUP_API_KEY, which is the fastest start when you already have an organization. If you do not have one yet, register the server with no credentials and sign in from chat instead — see Sign in from chat.
Cursor
Add this to your Cursor MCP configuration:Inside
craveup-turborepo, pnpm workspace resolution can shadow the published
package. If you are working in this repo, use the local repo build
configuration below.Codex
Add separate sandbox and production servers from the terminal:Claude Code
CraveUp is a local stdio MCP server, not a hosted HTTP endpoint. Register separate sandbox and production processes at user scope:/mcp in Claude Code to inspect or reconnect the server. Start a new Claude
Code session if the current session does not refresh its tool inventory after
registration.
Use -s project only when you intentionally want to commit a team-wide
.mcp.json. Never place API keys or access tokens in that file.
Claude Desktop
Add this to~/Library/Application Support/Claude/claude_desktop_config.json:
Local repo build
Use this when you are working insidecraveup-turborepo and need the newest MCP tools from source:
Environment and authentication
CRAVEUP_ENVIRONMENT is required and must be sandbox or production. The MCP server resolves the
dashboard, public/admin APIs, Clerk issuer/client, expected API-key prefix, and credential namespace
from that one value. It refuses a test/live key or endpoint mismatch before making a request.
The environment is immutable for the lifetime of an MCP process. Use two named
registrations rather than an in-session switch: craveup-sandbox for
development and craveup-production for go-live work. This keeps hosts,
credentials, API keys, and browser sessions isolated.
In sandbox, Crave uses its platform Stripe test account for PaymentIntents, so
developers can build a storefront and place test orders without creating a
merchant connected account or subscription. In production, the dashboard and
setup tools remain available before commercial onboarding is complete. Stripe
Connect KYC, banking, and subscription are required only before the merchant
can accept real payments.
The MCP server supports three authentication modes.
To create an API key, open Business Manager and go to Dashboard > Developers > API Development > Step 1 - API Keys. API keys are tied to an organization and may be restricted to specific locations; each MCP call still needs a target
locationId.
Sign in from chat
Requires@craveup/mcp 0.5.0 or later.
An API key is scoped to an organization that already exists, so it cannot unlock organization setup or the go-live tools. For those, the agent can start a browser sign-in for you — no craveup login in a terminal, and no API key in the configuration. CRAVEUP_ENVIRONMENT is still required, since it selects which dashboard and Clerk issuer the sign-in targets:
start_crave_auth— returns a hosted Crave Up URL. Open it and sign in, or create a free account.complete_crave_auth— finishes the exchange and stores the session.
~/.config/craveup/credentials.json, and is shared with the craveup CLI — so signing in through either one satisfies the other. Run whoami at any time to confirm which account is active.
On versions before 0.5.0 these tools are absent. Upgrade, or run
craveup login from a terminal to produce the same shared session.Normal MCP requests reuse the persisted CLI/MCP session; they do not open a
browser. Browser authentication happens only when
start_crave_auth (or
craveup login) is explicitly started. Sandbox and production sessions are
stored separately.Do not put production API keys in shared prompt text, screenshots, recordings,
or public repository files. Pass credentials through the MCP client
environment instead.
Available tools
Exact tool availability depends on the installed package version. The current repository source exposes these tool groups; installed npm versions may differ:
Multi-location and sandbox tools —
create_location, delete_location, list_locations, get_location_orders, list_customers, set_menu_active, and seed_sandbox_analytics — are available from 0.4.0. Browser auth is available from 0.5.0. The repo source can carry tools ahead of the published package; use the local repo build if you need one before its release.
For mobile releases, first run craveup apps configure in the Expo/React Native project and commit
the generated .eas/workflows files. The MCP release trigger only supports internal TestFlight. It
requires the exact confirmation phrase and a scoped Expo robot token because dispatching a workflow
consumes EAS build minutes. Android internal publishing stays in the CLI so the local Play submit
profile can be verified immediately before submission. Apple password/2FA prompts remain in a
visible terminal, and public App Store review remains a separate App Store Connect action.
Example prompts
Once connected, ask your agent for Crave-specific work:Faster setup with a prebuilt prompt
Use the prebuilt prompt when you want an AI agent to start with Crave Cloud, Crave.js, Storefront API, and MCP context already loaded.Use the storefront prompt
Copy a ready-to-use agent prompt for building a Crave-powered storefront.
Get an API key
Create the key and location ID your MCP server or storefront needs.