AI Frontier · Run in Agent
Claude Desktop — the platform, in an ordinary chat
Connect the platform's governed MCP server to Claude Desktop, then ask about your members, campaigns and metrics — or run an approved operation — right in a chat. No building, no dashboards to open. Every call uses the same tenant key, scopes, audit trail and budget as every other way you use the platform.
How it works
From connection to conversation in four steps
An MCP server is a governed connection your agent uses to call the platform's tools. Wire it up once, then just talk. See the MCP layer for the details.
Connect the MCP server
Add the platform's governed MCP server to Claude Desktop. An MCP server is simply a governed connection your agent uses to call the platform's tools — nothing is built or coded.
Authenticate with your tenant key
The connection signs in with your tenant key, so Claude is only ever talking to your workspace and your data — the same identity every other part of the platform uses.
The governed tools appear
Your approved tools show up inside the chat — each one carries the exact per-tool permissions (scopes) you've already set. Nothing outside those scopes is reachable.
Ask a question or run an approved operation
Ask about your members, campaigns and metrics in plain language, or trigger an operation you've approved. Read-only questions are open; anything that changes data is gated.
What you can do
Ask, act, and trust the number
Ask about your data
Who are my most active members this month? How did last week's campaign perform? Read-only questions are open and answered straight from your workspace.
Run an approved operation
Trigger an operation you've authorized — like sending an approved campaign — right from the chat. Anything that changes data or moves money is gated by scope and, where configured, approval.
Grounded in the semantic layer
Numbers come from the platform's certified semantic layer, so the figure Claude returns matches your dashboard. It never invents a number.
Governance
Same guardrails as everywhere else
Running the platform inside Claude Desktop doesn't open a side door. Every call inherits the controls you already trust — and it never trains on your data or acts on its own.
Tenant key
Every call is scoped to your workspace with the same tenant key the rest of the platform uses. Claude only ever sees your data.
Per-tool scopes
Each tool carries the exact permissions you've granted. If a scope isn't enabled, the tool simply isn't available in the chat.
Read vs. write
Read-only questions are open. Anything that changes data or moves money is gated by scope and, where configured, an approval step.
Audit trail
Every question and operation is written to the same audit trail as every other way you use the platform — nothing runs in the dark.
Budget
Usage draws on the same budget as the rest of the platform, so there's one place to see and control what your agent is doing.
Ask about your members, campaigns and metrics right in a chat.