Testing Your Approval Workflow¶

Editions: OSS, Cloud, Enterprise. Unless stated otherwise, everything on this page ships in OSS.

Verify your Preloop setup end-to-end: call a prelooped tool from your MCP client, receive the approval request, approve it, and confirm the tool executes and everything lands in the audit trail.

These tests use the hosted demo MCP server at https://example-mcp.preloop.ai/mcp, which provides harmless sample tools like pay and get_random_number.

Time Required: 10-15 minutes

Prerequisites:

Preloop account created (Creating Your Account)
API key generated
MCP client connected (Connect Your MCP Client)
At least one tool prelooped

Test 1: Basic Approval Workflow¶

Goal¶

Verify the complete approval flow works: call tool → receive notification → approve → tool executes.

Setup¶

Step 1: Add the Example MCP Server

If you haven't already:

Go to Tools > Add MCP server

Configure:

Server Name: Example MCP Server
Server URL: https://example-mcp.preloop.ai/mcp
Transport: HTTP Streaming
Authentication Type: None

Click Add
Click Scan for tools

Step 2: Preloop the pay Tool

Find the pay tool on the Tools page and expand it
Click Add rule and choose Require approval
Select an approval workflow, or create one with yourself as the approver
Leave the condition empty so every call needs approval
Click Add rule

Email notifications depend on your own settings under Settings > Notifications.

Execute Test¶

Step 1: Trigger Tool Call

In your MCP client (e.g., Claude Code):

Using @preloop tools, pay alice@example.com $500

Expected: the client sends the request and waits, the tool does NOT execute immediately.

Step 2: Check for Notification

Within 30 seconds:

Email: message from hello@preloop.ai, subject "Approval Required: pay", with the tool name, arguments, request ID, and Approve/Decline buttons
Web Dashboard: notification bell (top-right) shows a badge; click it to see the pending request
Mobile (if configured): push notification with tool name and key arguments

Step 3: Approve Request

Approve from any one channel:

Email: click Approve, confirmation page shows "Approval recorded successfully"
Web Dashboard: click the notification → review → Approve (optionally add a reason)
Mobile App: tap the notification → Approve → authenticate

Step 4: Verify Execution

In your MCP client the tool now executes and returns a result (illustrative):

Approval received
Payment of $500 sent to alice@example.com
Transaction ID: tx_abc123

In the dashboard, the request moves to History under Audit > Approvals.

Verify Audit Trail¶

Go to Audit > Approvals and look under History
Find your test request and click it
Verify it shows: requester, tool (pay), arguments, status (Approved), approver, timestamps, and the execution result

Result¶

Pass if: notification arrives within ~30 seconds, approval works, tool executes only after approval, and the audit log shows the complete record.

Fail if: no notification, tool executes immediately without approval, or no audit log entry.

Test 2: Decline Workflow¶

Goal¶

Verify declining an approval request prevents execution.

Execute Test¶

Step 1: Trigger Another Tool Call

Using @preloop tools, pay bob@example.com $1000

Step 2: Decline Request

When the notification arrives, click Decline (email or dashboard) and add a reason, e.g. "Test decline".

Step 3: Verify No Execution

Your MCP client receives a decline message (illustrative):

Approval declined
Reason: Test decline
The tool was not executed.

The dashboard shows the request as "Declined" with your reason, and no execution result.

Result¶

Pass if: the tool does NOT execute, the client receives the decline message, and the audit log shows "Declined".

Fail if: the tool executes despite decline.

Test 3: Conditional Approval¶

Goal¶

Verify CEL conditions work: approval only required when the condition matches.

Setup¶

Go to Tools, find the pay tool, and edit its Require approval rule
Set the condition to amount > 1000 with the condition builder, or as a CEL Expression:
```
args.amount > 1000
```
Click Update rule

This means: only require approval if amount > $1000.

Execute Test¶

Test A: Below Threshold

Using @preloop tools, pay charlie@example.com $500

Expected: no approval request, the tool executes immediately.

Test B: Above Threshold

Using @preloop tools, pay charlie@example.com $1500

Expected: approval request sent, client waits, tool executes after you approve.

Result¶

Pass if: small amounts execute immediately, large amounts require approval.

Fail if: all amounts require approval (condition ignored) or none do (condition broken).

Remaining Checks¶

Once tests 1-3 pass, the core Safety Layer works. Run through these quicker checks as needed:

Team-based approval with quorum (needs 2+ users): (Cloud and Enterprise) create a team under Settings > Teams, set it as the approver of the pay workflow with Approvals Required 2, and trigger a call. The tool must wait for the second approval before executing, and the audit log should show both approvers. See Team-Based Approvals.
Timeout and escalation (Cloud and Enterprise): set a short timeout (e.g. 120s) and escalation delay (e.g. 60s) on the pay policy, trigger a call, and don't respond. The escalation contact should be notified after the delay, and after the timeout the client should receive a timeout error with the tool NOT executed.
Multiple notification channels: enable email, web, Slack, and mobile for the pay policy and trigger a call. All configured channels should notify within ~30 seconds, and approving from one channel should update the others.
Non-prelooped tool: call get_random_number from the example server, which has no rule. It should execute immediately with no notification, but still appear in the audit trail.
Error handling: trigger a call with invalid arguments (e.g. pay invalid-email $-100) and approve it: the execution failure should be reported clearly. Then point the example server at an invalid URL and confirm you get a clean "server unavailable" error instead of a silent failure; restore the URL to https://example-mcp.preloop.ai/mcp afterwards.

Next Steps¶

Your setup is working. Next: Add your own MCP tools or create automated flows.