Eon cuts new-engineer tool setup from two days to two minutes

Eon, a cloud data protection company, says new engineers now reach all their investigation tools in about two minutes. Before, setup took two days. The fix was a single company-wide gateway on Amazon Bedrock AgentCore that replaced separate logins and local setups for each tool. More than 100 developers and AI agents use it across AWS, Google Cloud and Azure.

The problem

When a production problem hits, Eon's engineers check several systems in one session. A crashed Kubernetes pod, for example, means looking at container logs, Grafana metrics, the related GitHub pull request and the cloud provider's health status. Each tool had its own credentials and its own local program to connect AI assistants to it. As headcount and AI tools grew, keeping all that running became costly.

What Eon built

Eon weighed building its own single proxy with custom sign-in. That would have handled credentials but needed a lot of hand-built code for sessions and permissions. It chose a managed gateway from AWS instead. The gateway uses the Model Context Protocol (MCP), a common way for AI tools to plug into other software.

• Developers using Claude Code, Cursor or OpenAI Codex connect to one web address.
• They sign in once with their company login, which runs through Amazon Cognito.
• An autonomous AI agent for site reliability connects to the same gateway with its own identity.
• Hundreds of tools from dozens of servers appear in one list, with no local setup and no per-tool tokens.

The safety rules

The main security choice was to enforce limits at the gateway. Eon uses Cedar, a policy language, to spell out who can use which tool. The gateway checks every tool request against those rules before passing it on.

• Read access goes to selected groups, not every signed-in user.
• Production tools are read-only.
• Writes are limited to low-impact actions, such as commenting on a pull request or ticket.
• Write permission goes only to specific agent identities, never to human users.

Eon's example rule lets an agent named sre-bot comment on GitHub pull requests. Another rule blocks human users from calling any tool that changes things.

What owners can take from it

Many small businesses juggle a growing pile of tools, logins and now AI assistants. Eon's case shows the payoff of one controlled front door. New people get started faster. You also get one place to decide what staff and AI can see or change. The post is written by Eon with AWS, so the two-minute figure is the company's own claim.