Run your brand from your AI tools.

Your whole business is already behind one MCP endpoint.

Ask in Claude or Cursor and get the answer with the fix staged. You approve. It acts.

Live the same dayNo engineers required
Contribution margin this week
Curator, what is dragging contribution margin this week?
Thought for 2 seconds

I will pull margin drivers, shipping costs, and carrier emails through Curator.

CuratorChecking contribution margin1.6s
get_workspace_metrics(metric: "contribution_margin", window: "7d")
→ margin: 27.8% (prior week 31.2%, -3.4 pts)
→ driver: "shipping cost per order"
ShopifyReading shipping costs in Shopify2.1s
query_workspace_data(source: "shopify", dataset: "orders", field: "shipping_cost")
→ avg cost per order: $8.45 (was $6.10)
→ zone 4: 412 orders, 38% of volume
GmailSearching carrier emails3.2s
query_entities(source: "gmail", vendor: "Lunair Freight", topic: "rates")
→ Jun 28: zone 4 surcharge notice, +$2.35 per parcel
→ no volume discount since May
CuratorStaging the fix for approval1.2s
run_operation(operation: "stage_action", reroute: "zone 4 to East", dispute: "Lunair surcharge")
→ status: "staged", approval: "required", rollback: "available"
CuratorDrafting a report0.8s
present_dashboard(title: "Contribution margin investigation")
→ rendered: ui://curator/report
CuratorCuratorMCP app · ui://curator/report
Contribution margin · investigation
Margin · 7d
27.8%
vs 31.2% prior wk
Ship cost / order
$8.45
was $6.10 · +38%
Zone 4 orders
412
38% of volume
SUGGESTED

Reroute zone 4 orders to the Meridian East warehouse and draft a surcharge dispute with Lunair Freight. Estimated $4,900 per week back into margin.

routingzone_4 → "East warehouse" · 412 orders/wk
dispute"Lunair Freight" · Jun 28 surcharge
Approved · Reroute live, dispute drafted

The drop is shipping. Lunair Freight repriced zone 4 on Jun 28, taking average cost per order from $6.10 to $8.45 across 412 orders. The reroute and the dispute are staged in Curator and now approved.

Reply to Claude...
CuratorCurator
Claude can make mistakes. Please double-check responses.
Why it feels like an operator

The wiring is already done. Your agent just asks.

Reasoned answers, not raw rows
Reasoned answers, not raw rows

Your agent gets a resolved answer it can act on, not a table to parse.

One server, your whole stack
One server, your whole stack

Every tool you use is already behind it. Curator exposes the business as tools any client can call.

Rich output, inline
Rich output, inline

Curator renders dashboards and staged fixes right inside the conversation.

It learns your business
It learns your business

It remembers context across calls, so the answers get sharper over time.

How it works

Plug and play. Live in minutes.

No setup project. No IT ticket. No training videos.

Add the Curator MCP server in one line of config
1Add one endpoint

Drop the Curator server into Claude, Cursor, or your own agent. One line of config, nothing else to wire.

Your agent asks Curator a business question
2Ask a business question

Your agent calls Curator, which reasons over your live stack and returns an answer it can act on.

Approve a staged write from your agent
3Approve staged writes

Anything that changes a connected system comes back staged. A human confirms before it applies.

You stay in control

Full access for your agent,
full control for you.

Curator exposes read access by default. Anything that writes to a connected tool returns as a staged action a human approves, and every call is logged.

  • CuratorRead access open by default
  • CuratorWrites return as staged actions
  • CuratorEvery tool call logged and auditable
MCP: a staged action awaiting your approval

FAQ

It is a Model Context Protocol server that sits over the tools already connected to your Curator workspace. Any MCP-compatible client, like Claude or Cursor, can call it to query your business and stage changes, with Curator's reasoning layer in between.

You can wire up the pipes. The hard part is the reasoning: entity resolution across systems, joins, keeping data fresh, and handling APIs that quietly change. Curator does that work and keeps doing it, so your agent gets answers instead of raw rows.

Any MCP-compatible client, including Claude, Claude Code, and Cursor, plus your own agents built on the protocol.

Only with approval. Reads are open; any write returns as a staged action a human confirms, with one-click rollback and a full audit trail.

See all questions

Start with a free audit.
See it on your own data.

Connect your stack and Curator runs a full audit, free. Onboarding a limited number of brands this quarter, no credit card and no sales call to start.

See all plans
Curator
Free audit on your data
Every integration. See it before you commit.
  • Every integration in your stack
  • Slack, MCP, and the browser extension
  • Reports, dashboards, and alerts
  • Staged actions with one-click rollback