Securing Arcade MCP Deployments

You may have noticed that when you connected to the MCP serer you created with arcade-mcp, you could immediately call your tools from local MCP Clients and agents, like Claude and Cursor. This is because the arcade-mcp server is not secured by any mechanism by default. Most use-cases for MCP servers today are local development or local to a single machine, and we optimize for that use-case.

However, you can secure your MCP server by deploying it to Arcade (available today) or using OAuth (coming soon).

Arcade Deploy

When you arcade deploy your MCP server, it will be secured behind the Arcade platform.

Under the hood, we disable the MCP routes provided by arcade-mcp, and use the Arcade Engine as a gateway for your MCP server, which has a number of additional features. Arcade will create a randomized secure secret for your MCP server (via the ARCADE_WORKER_SECRET environment variable) so that your server is protected from unauthorized access, as well as being isolated from direct access from outside of the Arcade platform. Servers managed by Arcade (servers that are arcade deployed) serve /worker endpoints that are protected by this secret. The worker endpoints are worker/health, /worker/tools, and /worker/tools/invoke. The health endpoint is not protected by this secret, but the listing tools and tool invocations are. You can explore this locally by setting the same environment variable in your locally.

Learn more about how to deploy your MCP server to Arcade here.

OAuth (Coming soon)

Coming soon, you will be able to secure your MCP server’s /mcp endpoints with a OAuth Authorization Server (AS) - either using Dynamic Client Registration (DCR) or Client ID Metadata Documents (CIMD). Learn more about how MCP integrates with OAuth here .