How Ask Calm and Agent MCP work
CalmCompliance has two AI answer surfaces: Ask Calm, a conversation inside the Command Palette on the site you have open, and Agent MCP, which connects an external AI client such as ChatGPT, Claude, Copilot, or Cursor to Calm. Both answer questions about tickets and work orders using the same five read-only reporting tools, and both show only what the person asking is allowed to see.
Ask Calm
Ask Calm is a conversation inside the Command Palette overlay, not a separate page or chat window. Open the palette on a site page with Command+K or Control+K, then choose Ask Calm, or type a question and pick the Ask option. It answers about the site you are currently viewing, using your signed-in access.
Ask Calm appears only when AI features and the Command Palette are enabled for your organisation. The first time anyone in your organisation opens it, the palette shows a consent panel headed "Ask About" followed by your site's name. The panel states: "Answers are produced by sending relevant site information to Calm's AI provider. Models are not trained on your organisation's data." Your choice is remembered on that device for the organisation, so the panel does not appear again on that device.
Agent MCP
Agent MCP connects an external AI client to Calm, so you can ask the same questions from ChatGPT, Claude, Copilot, or Cursor. The settings page is titled "Agent MCP", with the subtitle "Connect ChatGPT, Claude, Copilot or Cursor to Calm". It lives in Calm Settings > Agent MCP and has About, Setup, and Audit tabs; the client setup instructions are on the Setup tab.
A connection uses OAuth 2.1: you add the MCP URL as a remote server in your AI client, then sign in to Calm as yourself and allow the organisation, sites, and read access. Write access is optional, and we do not issue API keys for MCP. The organisation must allow MCP connections, and signing in through impersonation or view-as cannot complete a connection.
The five reporting tools
Both surfaces share the same read-only reporting tools for tickets and work orders. They only read records; they never create or change anything.
get_ticket_metrics: complete ticket counts, current open age, and breakdowns by status, severity for issues, priority for requests, who raised the ticket, and assignee.get_work_order_metrics: complete work-order counts with breakdowns by status, priority, and active assignment to an individual or a team.get_ticket_category_metrics: complete visible ticket counts by stored category, with creation-window and open/closed filters.get_status_history: paged recorded status transitions and elapsed hours for an issue, request, or work order.get_closure_metrics: time-to-close aggregates for issues, requests, and work orders, including closure windows, outcomes, and missing-date counts.
The module-specific tools are advertised only when the relevant Service Desk or Work Orders capability is available.
What the answers can see
Results follow the normal access rules of the person asking, such as the Service Desk roles and access that control who sees tickets. Ask Calm answers about your current site using your signed-in access. Agent MCP answers use the sites granted to the connection and the connected person's remaining access. Records the caller cannot see are treated as absent: anonymous reporters, and location or asset names hidden by permissions, stay masked.
Where to go next
To run a report in Calm itself, follow Use Ask Calm for ticket and work-order reports. To work from a connected AI client, see Ticket and work-order reporting with Agent MCP. For how the counts are calculated and why two people can get different numbers, read How AI report counts and visibility work.