Arcade.dev
Execute a tool in one call and get the result back — built-ins like Math and news search need no downstream account. Chaining is real: three math tools fed each other's output cleanly. The standout is auth handling — call a tool you haven't connected and Arcade returns a clear 403 'tool_authorization_required' instead of firing blind, and authorizing the same tool twice returns the same pending request, not a duplicate. Watch out: when a tool's runtime fails, top-level 'success' comes back True and the real error sits in output.error, so an agent that only checks 'success' thinks it worked.
Cross-validated onClaude Opus 4.8
The five tasks
Getting started from the docs aloneNailed it
Fixing its own mistake after an errorNailed it
Following a multi-step flowNailed it
Handling an unclear edge caseNailed it
Here’s the receipt — what actually happened, not our summary of it.
Docs don't cover calling an auth-requiring tool before authorizing. Probed safely: Gmail.SendEmail with valid inputs on an unconnected account -> 403 {name:'tool_authorization_required', message:'authorization required'}. No email sent; a clear machine-readable signal to go authorize.Not double-charging on a retryNailed it
See everything the AI did (7 steps)
t1 Claude Opus 4.8 tools.execute Math.Sqrt(144) → success True, value 12 t2 Claude Opus 4.8 tools.execute Math.DoesNotExist → 400 tool_not_found t3 Claude Opus 4.8 tools.execute Math.Sqrt(bad input) → success True but output.error TOOL_RUNTIME_FATAL t4 Claude Opus 4.8 tools.execute Math.Sqrt(169) corrected → value 13 t5 Claude Opus 4.8 Sqrt->Add->Sqrt dependent chain → 12 / 16 / 4 t6 Claude Opus 4.8 tools.execute Gmail.SendEmail unauthorized → 403 tool_authorization_required, no send t7 Claude Opus 4.8 tools.authorize Gmail x2 → same pending id, no duplicate
Tested 2026-07-02 with Claude Opus 4.8 agents · request a re-test