Why a Sandbox
Configuration errors surface fast when the sandbox and production share the same transport and permission model. The v2 sandbox mirrors your server's stdio and HTTP entry points so a passing smoke test means a passing deploy.
Every session starts clean: no lingering state, no shared ports, no leftover processes. You can reproduce a flaky agent failure and inspect it before it costs real requests.
Pro Tip: Mock the Edge Cases
Configure the mock transport to simulate slow responses and timeouts. Most agent failures are latency failures, not logic failures.
Launching the Sandbox
# Launch the sandbox server with the mock transport
npx toolverse sandbox --port 4000 --mock
# Register tools, then smoke test against them
npx toolverse sandbox:test ./server.mjs