AI NewsInfrastructureAnnouncement

Zalando opened an OAuth broker that gives AI agents scoped sign-in

Zalando released an open-source broker. An AI agent signs in to Google, GitHub or Linear through it. The user can take the access back.

AI News

Editorial2 min read

LinkedInX
GitHub social card for zalando-incubator/agentic-identity-broker

Image: GitHub

Why it mattersAn AI agent that reads your mail or files is a new kind of login. The broker gives you one place to set and remove its permissions.

An AI agent that reads your email or your Google Drive is a new kind of login. The common answer has been to hand the agent a password or a long-lived token. Both choices give up control. A new open-source project from Zalando tries to do better.

Zalando Incubator released Agentic Identity Broker this week. The project is a server, written in Go, under the MIT licence. A Show HN post has 19 points on Hacker News at the time of writing. The repository has 52 stars. The README says Zalando uses it in production for several of its internal MCP servers.

What the broker does

The user signs in to the broker once. The broker then holds the tokens for each outside service. When an AI agent wants to call Google, GitHub, Databricks or Linear, the agent goes through the broker. The broker exchanges its own token for a short, scoped token for that service. Every call is written to a JSON audit log.

Before any new access, the user sees a consent screen in the broker. The screen says which agent is asking, which service it wants to reach, and which scopes it will use. The user can say no. The user can also take an access away later. A separate tool approval feature can block a call just before it runs. The check works on an agent of any kind.

The broker also supports delegation chains. One agent can call another on behalf of a user, with the scopes it has been given. A privileged bridge, like a company Slack, can start a chain for a user who is not in the room. The team says this delegation is part of the design.

What the project does not do yet

The roadmap in the README names three jobs still to come. Formal token exchange between agents. Authorization based on what the agent intends to do, so the check can go beyond the OAuth scope. And an activity log the user can read directly.

The team writes that the project is "in active development". The maintainers, according to the README, have built several OIDC and OAuth2 providers before this. The project sits inside the zalando-incubator GitHub organisation.

A team building its first AI agent on company data can wait on a broker like this. A team running a second or third agent cannot. The broker moves the password out of the agent. It puts it in one place, with a record.

Source

SourceGitHub

This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.

Share
LinkedInX
Start a project