Supervaize MCP: example usage
Once the connector is installed, talk to your workspace in plain language. Pick a conversation below and watch which tools the assistant calls.
Illustrative replay with fictional data. Wording varies between chats; the tools are the real ones.
Prompts to try
Copy any of these into Claude or ChatGPT.
See what is going on (read)
- What is running in my Supervaize workspace?
- Which missions have active jobs, and who is assigned to them?
- How did the Customer Voice mission do over the last 30 days? Cases by status and total cost.
- Show me the steps of the last failed case and tell me where it went wrong.
- Which agents are online?
Unblock an agent (write)
- Is any agent waiting on me?
- Show me the question on that case, then approve it.
The assistant reads the pending human-in-the-loop step, shows you the question, and submits your answer with step_submit. The agent resumes at once.
Launch work (write)
- Create a mission called "Q4 Check-ins" and assign Aidan to it.
- Start a new job for Aidan in the Onboarding Check-ins mission.
Creating a job is always two-phase: the assistant first calls job_setup in preview mode and shows what it would create. Nothing starts until you confirm.
:::tip Write a weekly report in one prompt
Every Monday I want a status of all in-progress missions: cases completed last week, cost, and anything awaiting input. Write it as a short email to my team. The assistant chains mission_list, mission_stats and case_list on its own.
:::
How a request flows
Vocabulary
| Term | Meaning |
|---|---|
| Mission | Groups the work of one or more agents |
| Job | One agent run inside a mission |
| Case | The agent's unit of work inside a job, for example one interview or one ticket |
| Step | One ordered event of a case; some steps wait for a human answer |
Tool reference
The connector exposes 18 tools. You never call them yourself; the assistant picks them. Write tools need the Write permission granted at connection time and the matching role in Studio.
| Tool | What it does | Access |
|---|---|---|
workspace_info | Connected workspace, your roles, granted permissions | Read |
agent_list | Agents in the workspace and their status | Read |
catalog | Data resources and analytics each agent declares | Read |
mission_list / mission_get | Browse the missions you can see | Read |
mission_stats | Jobs and cases by status, cost, activity per day | Read |
job_list / job_get | Browse jobs, their cases and the agent's actions | Read |
case_list / case_get | Browse cases and their ordered steps | Read |
resource_list / resource_get | Items of an agent-declared data resource (sensitive fields masked) | Read |
artifact_get | Fetch a deliverable attached to a step | Read |
mission_create | Create a mission, optionally assigning agents | Write |
job_setup | Create a job for an agent: preview, then confirm | Write |
step_submit | Answer a human-in-the-loop step | Write |
agent_action_invoke | Run an action the agent declares on a job | Write |
job_sync | Reconcile a job with its agent after a network hiccup | Write |
Good to know
- Sensitive fields stay masked in data resources unless your role can configure agent parameters.
- Every call is logged with the tool name and who made it. Argument values and responses are not stored.
- Sessions last. The assistant refreshes its access silently; after 30 days without use you sign in again.