letsauth

About

Let's Auth Labs

An independent, vendor-neutral lab for how AI agents are authenticated and authorized. We measure what the protocols require and what real systems do, and publish all of it as open data.

Why a lab

Agents now call APIs, use tools and act for people. The rules for how they get and pass on authority sit where agent protocols (MCP, A2A) meet identity standards (OAuth, OpenID Connect, token exchange, workload identity), and both sides move monthly. Most of what is written about it is assertion. Let's Auth Labs replaces assertion with evidence: every claim cites a specification revision, every status cites a dated observation, and the data is yours to download and check.

What we build

  • MCP auth atlasLive

    Public MCP servers against every requirement of the MCP Authorization specification, one cited observation per cell.

  • Agent auth atlasLive

    The specifications, drafts and EU acts behind agent authorization as one map of what builds on and competes with what.

  • ScannerLive

    A nightly job that reads listed servers' public metadata and proposes new observations as pull requests.

  • Delegation chain explorerPlanned

    Step-through scenarios with the decoded token at each hop, and toggles for the attacks that break them.

  • Breach anatomyPlanned

    Each agent-auth vulnerability replayed as a scenario, with a lab to reproduce it.

  • Agents under EU rulesPlanned

    What eIDAS 2.0 and the EUDI wallet, the AI Act and NIS2 ask of agent delegation, as scenarios and long reads.

All of them are views on one shared data model. The order and status of each is in the roadmap.

How we work

  • Cited, not asserted. A claim without a specification revision and section does not ship.
  • Neutral. Nothing is ranked or scored. Rows are alphabetical, and no vendor pays for anything.
  • Unknown is honest. A missing observation reads unknown, never failure.
  • Passive by default. We read public metadata only, on a schedule, never on a visitor's request, and never with credentials.
  • Open. Code under MIT, data under CC BY 4.0, history kept in a public repository where anyone can propose a correction.
  • European by default. Built in Berlin, hosted and measured in the EU; see made in Europe.

Who is behind it

Let's Auth Labs is run by Jan Brennenstuhl, a security engineer in Berlin who works on usable security and web authentication. It is a personal, independent project; the disclaimer says what that means for affiliation.

Get involved

Know how a server behaves? Add an observation. Found a mistake, or want a specification on the map? Open an issue in the repository. Anything else: letsauth@janbrennenstuhl.eu.

Last updated