MCP tool groups and capability discovery
Discover current inputs, permissions, cost classes and review requirements rather than relying on an old count.
Use tools/list to discover the current connector. A count does not prove a particular app action, filter or recovery step is supported.
The developer catalogue comes from the release registry and includes complete input schemas. For newer customer operations, customer_operation_catalogue reports consent scope, cost class, human review and execution availability.
| Area | Examples |
|---|---|
| Business book and searches | Search filters, records, saved lists and searches. |
| Agents | Settings, jobs, context, drafts, decisions and tasks. |
| Email outreach | Campaigns, sequences, recipients, replies and sending controls. |
| LinkedIn and posts | Campaign work, authored posts, media and manual activity review. |
| Calling | Lists, assignment, next record and saved outcomes within the user's access. |
| CRM | Records, fields, tasks, notes, views, setup and imports. |
| Workflows and recovery | Read, test, prepare publishing or restore, then fetch the result. |
| Account, team and reports | Preferences, permitted team changes, report schedules and connections. |
| Google Drive backups | Connection status and run history, human Google consent and protected on-demand backup or disconnect, when registered. No customer backup schedule or Drive restore operation. See Google Drive backup through MCP. |
Reads return existing information. Ordinary management can execute with separate consent and validation. Protected operations prepare an exact human review. Specialist jobs explicitly request agent work.
Older write scopes do not grant new management permissions. Account roles still apply. Explicit model, research and external work can cost resources. Unavailable execution must not be treated as success or bypassed.
Read saved results without creating unnecessary agent jobs. See How protected MCP actions reach you for approval and MCP connector tools.