Observe

Playground

Test a Tool Pack interactively without standing up an agent.

The playground is a chat interface inside the Agent Handler dashboard. You pick a Tool Pack, pick a Registered User, and start typing - the agent responds with tool calls streaming inline. Every call hits the same MCP endpoint your production agent will use, so what you see in the playground is what your agent will see in production.

Use it when you’re tuning a Tool Pack - confirming the model picks the right tools, that descriptions read well, that input overrides constrain correctly. Tighter feedback loop than running the full agent against a test prompt.

The playground is currently feature-flagged. If you don’t see it in the sidebar, ask your account team to enable it.

Opening the playground

Playground in the sidebar. Three things to set up before you can chat:

  1. Pick a Tool Pack from the dropdown. The playground shows the tools in the pack so you know what’s available.
  2. Pick a Registered User. Either an existing one (production or test) or a freshly minted test user.
  3. Make sure Connectors are authenticated. If the Registered User hasn’t authenticated a Connector the Tool Pack expects, you’ll see a “Connect [Connector]” button - click it and run through Link.

Once those are set, type a message and the agent runs.

What you see when the agent runs

Each tool call shows inline in the chat:

  • The tool name (slack__post_message).
  • The arguments the model produced.
  • The result.
  • The latency.
  • Any redactions the Security Gateway applied.

If a call fails, the error class and message are shown in line. If a Security Gateway rule blocks a call, the violation reason is shown - you can verify your rules are firing as expected.

Common workflows

Tuning tool descriptions. Ask the agent to do something a specific tool should handle. If it picks the right tool, the description works. If it picks something else (or asks a clarifying question), the description needs work - open the Tool Description Override editor in another tab, edit, save, and re-run the prompt. Changes apply on the next tools/list, which the playground refreshes between turns.

Testing input overrides. Force a field to a fixed value, then ask the agent to do something that should call the tool with that field. Check in the playground that the call arguments contain the forced value - that confirms the override applied.

Testing security rules. Construct a prompt that should trigger a rule (an email address, a credit card number, an internal customer ID). Run it. Watch for the redaction or block in the playground, then cross-check on the Alerts dashboard.

Reproducing a customer issue. Pick the production Registered User that hit the problem, copy the prompt from your support ticket into the playground, run it. You’ll see exactly what their agent saw, and you can iterate on the fix without bothering them.

Deep-linking to your IDE

The playground can hand off to Cursor or VS Code with the current context preserved. The Open in IDE button generates a deep link that pre-fills your local agent with the same Tool Pack and Registered User.

Use this when you want to:

  • Move from quick exploration in the playground to writing actual prompts in your codebase.
  • Reproduce a complex flow across multiple turns without the playground’s chat memory.
  • Pair-program with your agent on the same context the playground was running.

Pick Cursor or VS Code from the dropdown; copy the deep link or click to launch your IDE directly. Your IDE’s MCP config gets updated to point at the same MCP URL the playground was using.

Limitations

  • No multi-turn context export. Conversation history stays in the playground. The deep-link hands off the configuration (Tool Pack and Registered User), not the chat transcript.
  • No agent code customization. The playground runs a default agent loop. If your production agent has a custom system prompt or non-default model parameters, the playground’s behavior may not match exactly. Use the playground for tool selection and tool call shapes; use your own agent runtime for prompt-engineering work.
  • Production rate limits apply. Calls against a production Registered User count toward your monthly tool-call quota. For heavy testing, use a test Registered User.

Next

Manage who on your team has access to what with Team and roles.