Arcade AI

Arcade AI is infrastructure for production AI agents, not a general chatbot builder. The fetched homepage calls Arcade “The MCP Runtime for Production AI Agents” and describes a runtime for secure agent authorization, reliable tools, and governance. That positioning is specific: Arcade focuses on letting agents take authenticated actions in real services while preserving identity, permissions, and operational control.

The fetched pricing page was concrete. Hobby is free and includes 100 user challenges into services, 1,000 standard tool executions, 50 pro tool executions, MCP compatibility, one Arcade-hosted MCP server, and community support through GitHub. Growth is listed at $25 per month plus usage, including 600 user challenges before $0.05 overages, 2,000 standard tool executions before $0.01 overages, and 100 pro tool executions included. Enterprise is for larger deployments with custom requirements. These numbers make Arcade easier to budget than many agent-infrastructure tools, but usage still needs monitoring once agents run many authenticated actions.

Arcade should be compared with Model Context Protocol, Zapier MCP, Composio, Browserbase, and Cloudflare AI Gateway. MCP defines the protocol layer, Zapier MCP exposes a broad no-code app ecosystem, Composio focuses on authenticated integrations, Browserbase handles reliable browser execution, and Cloudflare AI Gateway helps with model routing and observability. Arcade’s edge is combining MCP hosting/runtime ideas with authorization and governance for production agents.

Use Arcade when your team is already building agents and needs controlled SaaS actions: Gmail, Slack, Salesforce, internal tools, or other APIs where user identity and permissions matter. It is not the right first choice for a non-technical team that only wants simple automations. A good pilot should define which tools the agent can call, which actions require approval, what logs are retained, and how revocation works. Track tool-execution volume, authorization failures, user challenges, and incidents. The upside is safer agent action; the cost is engineering implementation discipline.

Implementation checklist: design the permission model before writing agent code. List each tool, action, user challenge, approval requirement, and audit event. Start with low-risk read actions, then add writes such as Slack messages, CRM updates, or email drafts only after authorization flows are tested. Monitor standard and pro tool executions separately because usage-based pricing can surprise teams when agents loop or retry. Decide how users revoke access and how developers debug failed tool calls without exposing private data. Arcade is valuable when production agents need safe authenticated action; it is unnecessary overhead for a prototype that only calls one internal API.