Agent Sandbox — test MCP and UCP queries live

Last updated: April 2026

The Agent Sandbox at /dashboard/settings/agents lets you run MCP and UCP queries against your OWN live store, with the same tool calls and rate limits that production AI agents hit. Useful before you list on an AI Shopping channel or when a buyer reports an issue with an agent response.

Pick a persona

  • Gemini Shopping — Google's shopping agent
  • ChatGPT Commerce — OpenAI's commerce tool
  • Perplexity Shop — Perplexity's shopping surface
  • Claude Buyer — Claude running against your MCP endpoint

Run a query

  1. Go to /dashboard/settings/agents
  2. Pick a persona from the dropdown
  3. Type a buyer-style question ("Do you have any iPhone 15 Pros in good condition under $800?")
  4. Hit Send — the agent's tool calls and responses stream inline

Available tool calls

  • inventory.search — search your catalog by model, grade, price, storage
  • orders.summary — high-level order stats (for agents that handle post-sale)
  • warehouses.list — locations and stock per warehouse
  • check_imei — lookup a specific IMEI in your inventory

Why this matters

  • Validate your catalog data (names, prices, stock) before AI channels surface it to buyers
  • Reproduce customer-reported agent issues to debug them
  • Explore what the MCP protocol is actually capable of, without any external setup
NOTE

The sandbox uses your real, live data — not mock. Rate limits match prod, so heavy back-to-back queries can hit the same ceilings a production agent would.

TIP

If an agent returns a wrong or missing answer, run the same query in the sandbox — you'll see the exact tool call and response the agent used to build its reply.

Was this article helpful?

Back to Integrations