MCP Integration
Quick summary of Model Context Protocol integration
MCP Integration
The Model Context Protocol (MCP) gives your AI assistant direct access to your data sources (Database, Figma, Codebase).
Detailed configuration guides are linked per service below. For a step-by-step tutorial on what MCP is, check the Learning Center.
Supported Services
| Service | Feature |
|---|---|
| Supabase | Schema Inspection, SQL Generation |
| Firebase | Project setup, Firestore, Auth, messaging, Crashlytics, and Remote Config |
| Wire AI | Ordinary product events, onboarding insights, experiments, and learning reports |
| RevenueCat | Subscription Management |
| Figma | Design-to-Code |
| Context7 | Semantic Codebase Search |
Check the Learning Center for best practices and security.
MCP in the Boilerplate
The reviewed AI Pro and Standard candidates do not ship a .mcp.json. Configure only the servers you need in the active coding host, then record the real connection evidence in docs/KEYS.md.
An MCP setup choice is not a connection receipt. Keep keys in local environment or host secret settings, never in generated planning documents or git.
Firebase MCP → Firebase project operations
The official Firebase MCP server can work with project configuration, Firestore, Authentication, messaging, Crashlytics, Remote Config, and other tools exposed by the Firebase CLI. Its current tool list does not include Google Analytics queries. Ordinary app, screen, onboarding, purchase-funnel, and activation events use Wire by default and should be queried through Wire instead.
"List the Firebase apps registered in the active project."
See the Firebase MCP guide for the official CLI setup and exact config block.
Wire AI MCP → your product and onboarding insights
Wire AI runs a hosted MCP server, the same model as Supabase or HeyGen. Nothing runs from your repo: no server to build, no self-host, no local config file. You point your AI assistant at the hosted URL and authenticate with your key.
- Server URL:
https://wireai-mcp.fly.dev/mcp(HTTP transport) - Auth: an
Authorization: Bearer <key>header on every request
Add it to Claude Code with one command:
claude mcp add --transport http wireai https://wireai-mcp.fly.dev/mcp \
--header "Authorization: Bearer <key>"For any other MCP client (Cursor, Codex, and the rest), paste the same URL and bearer header into that client's MCP settings.
Two key scopes. Which tools you get depends on which key you authenticate with:
| Key | What it unlocks |
|---|---|
Tenant wai_ key | Read + loop tools (wireai_get_insights, wireai_submit_learning_report, wireai_mark_learning_applied) plus the no-auth guide/components tools |
| Admin key | Everything above, plus app management (register / update / list) and experiment tools (wireai_get_experiments and the three arm-setters for onboarding / review / questionnaire arms) |
The backend enforces scope per call. Grab your key from the getwireai dashboard: open your app, go to the Set Up tab, copy its key. Read it from your environment; never commit it.
Download insights as .md for your AI
Prefer to hand the data to an assistant that isn't wired to MCP? Wire AI exposes the same insights as a plain markdown file you can download and paste in:
- Via the API:
GET /v1/me/apps/{appId}/insights.mdreturns the analytics as markdown. - Via the getwireai dashboard: a one-click download
insights.mdaction.
The .md format is deliberate: it drops straight into any AI chat as context, so your assistant can reason over real onboarding behaviour when you ask it what to build or change next.