@cross-deck/ai

Crossdeck AI

Connect Crossdeck to Claude, Cursor, or any MCP client and ask your app questions in plain English — and get rendered charts and dashboards back, not just numbers. "Why did revenue drop?" "Who did this crash affect — do they pay us?" "Draw user growth over the last 30 days."

Crossdeck joins the layers about your users — identity, revenue, entitlements, errors, analytics, and database read-cost — into one place, by identity. Crossdeck AI gives that source-of-truth a voice. MCP is the transport; the product is Crossdeck understands your app, in whatever AI tool you use.

The point isn't "let AI read my data." It's the crossing: because Crossdeck owns the identity that ties the layers together, one question can span them — who an error hit and how many of them pay you. No single-layer analytics or error tool can answer that.

Test it in 10 minutes

1. Connect

Directory / remote (recommended): add the Crossdeck connector in your client's connector settings and complete the one-click sign-in — you'll authorize Crossdeck and pick a project. No keys to paste. (OAuth 2.0.)

Local (Claude Desktop / Cursor / Claude Code): install and pass a secret key from your Crossdeck dashboard → API keys:

npm install -g @cross-deck/ai

{
  "mcpServers": {
    "crossdeck": { "command": "crossdeck-ai", "env": { "CROSSDECK_SECRET_KEY": "cd_sk_live_…" } }
  }
}

Use a secret key (cd_sk_), never a publishable (cd_pub_) one — these are server-side reads of private data. CROSSDECK_API_BASE overrides the endpoint (e.g. sandbox).

2. Try these prompts

Ask	Tool(s) used	Expected outcome
"What's our MRR and how many paying customers, split by Stripe/Apple/Google?"	get_revenue	Current MRR in dollars, paying-customer count, and the three-rail split.
"Draw our user growth over the last 30 days."	draw_user_growth	A rendered line chart of unique visitors + page views over time, with totals.
"This error a1b2c3 — who did it affect, and do any of them pay us?"	get_error_impact	The issue's type/status/occurrences plus affected users and how many are paying.
"Show me everything about customer agent_8842."	get_customer → moat dashboard	A rendered cross-match: what they pay × entitlements × read-cost, in one view.
"What's driving our database reads — per-user vs overhead?"	get_read_cost	The per-user-vs-overhead split + reads by operation.

3. What you'll see

Data tools return clean JSON; the draw_user_growth and open_moat_dashboard tools render interactive charts/dashboards inline (in hosts that support MCP Apps; they fall back to a text summary elsewhere).

Tools

Tool	Title	Answers
get_revenue	Get revenue	MRR / paying count / per-rail / trend
get_read_cost	Get database read-cost	per-user vs overhead, by operation
get_error_impact	Get who an error affected	who it hit + how many pay
get_customer	Get a customer's full picture	revenue × entitlements × read-cost
get_host_analytics	Get analytics for a host	per-subdomain views + uniques
get_host_top_pages	Get top pages/referrers for a host	per-host breakdown
draw_user_growth	Draw user growth over time	rendered chart
open_moat_dashboard	Open the cross-layer dashboard	rendered cross-match

Every tool is read-only (readOnlyHint), has a human-readable title, returns scoped/paginated results, and surfaces actionable errors. All reads are point-reads of maintained ledgers — asking questions never runs up your database bill.

Known limitations

• Read-only in v1 — no writes, no config changes (those are a later, separately-gated stage).
• Per-host analytics require the host to be a verified origin of your project.
• Rendered charts require an MCP-Apps-capable host; other hosts get a text summary.
• Reporting reflects data going forward from when Crossdeck was connected (historical backfill is separate).

Privacy

Crossdeck AI reads only your own project's aggregate data, scoped by your token. It does not read your conversation history or local files. Full policy: https://cross-deck.com/legal/privacy/.

Support

[e-mail] · docs: https://cross-deck.com/docs/reporting-api/

License

MIT.