• FAULT
  • REACTION
  • HEALTH
  • HUMAN

How to simulate API failure scenarios with your AI

Connect Reqres to your agent to inject faults into the sandbox. You can then watch how your agent reacts when services go dark.

Ask AI about this page

Short answer

How can I test my AI against API failures?

Use the Reqres connector to trigger specific error codes in the agent sandbox. This lets you verify if your agent recovers or fails when a service becomes unreachable. You'll see exactly how it handles timeouts and broken responses.

Testing outcomes.

Where the results land.

Injecting faults reveals how your agent behaves under pressure.

FAULT

Triggered errors

The agent sandbox receives the specific error code you requested. This mimics a real-world service disruption.

REACTION

Agent behavior

Your AI attempts to process the failure. You can see if it retries the request or reports the error correctly.

HEALTH

System status

The agent's operational state is checked after the disruption. This confirms the agent hasn't crashed during the fault.

HUMAN

Manual intervention

If the agent fails to recover, you'll need to step in. This outcome flags when your logic needs a human fix.

The workflow

What your AI does when the failure is triggered.

The agent moves from injecting the fault to evaluating its own stability.

  1. Inject the fault

    You tell the agent to simulate a specific error condition in the sandbox.

    trigger_scenario

  2. Run the test

    The agent executes the requested scenarios to see how the environment reacts.

    agent_scenarios

  3. Check stability

    The agent monitors its own status to ensure the error didn't break its core logic.

    agent_health

  4. Report results

    The agent summarizes the outcome of the disruption test.

    agent_health

Try it

Copy these to start.

Use these prompts to begin testing your agent's resilience.

Starting points

Swap out the error types and scenario names to match your specific test case.

Reqres Connector

You're all set. Choose your MCP client and follow the setup instructions.

Connector linkhttps://edge.vinkius.com/vk_preview_AtxnuBtQ26tiOA6pBRuixOGykiHBrl8ytWq6813r/mcp

Claude Desktop

Follow the steps below to connect in seconds.

  1. 1In Claude Desktop, open Settings → Connectors.
  2. 2Click “Add custom connector” and paste the connector link above as the remote MCP server URL.
  3. 3Click Add and start a new chat — Reqres capabilities are ready to use.
Configuration · claude_desktop_config.jsonCopy
{
  "mcpServers": {
    "reqres-mcp": {
      "url": "https://edge.vinkius.com/vk_preview_AtxnuBtQ26tiOA6pBRuixOGykiHBrl8ytWq6813r/mcp"
    }
  }
}
Copy into chat04

  • Trigger a 500 error scenario in the Reqres sandbox.

  • Run the timeout scenario and tell me how you handle it.

  • Check my agent health after I trigger a connection failure.

  • Simulate a 404 error and show me your error handling logic.

  • Claude
  • ChatGPT
  • Cursor
  • VS Code
  • Windsurf
  • Claude Code
  • JetBrains
  • Cline

Start here

Connect Reqres once, then ask.

One connection gives your agent access to the sandbox. You can find the full list of available scenarios on the Connector page.

Connect Reqres to your AI

FAQ

How this task behaves.

  • 01

    Can the AI fix the API error itself?

    No. The AI can only react to the error you've injected into the sandbox.

  • 02

    Does this affect my real production data?

    No. These scenarios run within the isolated agent sandbox.

  • 03

    Can the AI delete my existing API configurations?

    No. The tools only allow for triggering scenarios and checking health.

  • 04

    Can I use this to test my own custom API?

    No. This connector is specifically for simulating failures via Reqres scenarios.

  • 05

    Will the AI automatically retry every failed request?

    The AI will follow whatever logic you have programmed into it, but the connector itself doesn't force retries.

  • More questions about Reqres? The Connector page answers them. See everything the Reqres Connector can do

Connect Reqres to Claude, Cursor, ChatGPT & more