1. Pick a pack
Open /marketplace and browse by category, vertical, or capability. Every listing has a trust card with the eval score, popularity, last update, seller reputation, and required connector scopes. Click into the detail page to read the ROI hypothesis and inspect sample inputs and outputs before you commit.
Tier eligibility is shown inline. If your current plan does not cover a paid pack, the install button is disabled and the price is displayed in plain USD with the per-run or per-month rate.
2. Choose an approval policy
Approval is the most important thing you configure at install time. Three presets cover almost every use case:
- Auto — the agent runs end-to-end without a human checkpoint. Use for read-only or low-risk packs.
- Approve — every external write, send, delete, or charge pauses for a named human to review the hash-pinned preview.
- Approve dual — two distinct humans must sign off; the requester is barred from being one of them. Recommended for irreversible or high-spend actions.
You can adjust the policy per tool after install. See /docs/permissions-and-approvals for the full enforcement model.
3. Authorize connectors
Most packs need at least one connector — a CRM, a chat system, a file store. The install flow surfaces the exact OAuth scopes the pack will request; pick the narrowest set that covers your use case. If a connector is already authorized in your workspace, the pack reuses it instead of asking again.
4. Run the bundled sample suite
Every marketplace pack ships with a sample-IO suite. After install, click Run sample on the agent page; the agent executes against the bundled inputs in a sandboxed dry-run. Review the outputs, citations, and timing before pointing the agent at real data.
5. Run on real data
Once the sample run looks right, kick off a real run from the dashboard or call the API:
# Start a run from the API
curl -X POST -H "Authorization: Bearer $TOKEN" \
-H "Content-Type: application/json" \
-d '{"input": {"customer_id": "cus_123"}}' \
"https://app.mvsagents.ai/api/v1/orgs/$ORG_ID/agents/$AGENT_ID/runs"Watch the live timeline from the agent page; approvals route to your approver in real time. Replay any historical run to debug or to verify a new agent version against your own traffic — see /docs/run-replay.
Uninstalling
Open the agent page and click Uninstall. Past run history is retained per the audit-log retention window so post-uninstall investigations and compliance evidence packs still work; see /docs/audit-logs.