Delegation chain explorer
Watch authority travel from a person to an agent to an API, one hop at a time
Each scenario is a chain of requests between a user, agents, MCP servers and authorization servers. Every step cites the specification section it follows and shows the token it carries, decoded, with the claims that changed since the previous hop. Turn an attack or a misconfiguration on to see which steps change and what the token looks like once a control is gone.
- Scenarios
- 3
- Steps
- 36
- Toggles
- 8
- Specifications cited
- 9
An MCP client obtains a token for an MCP server
The baseline chain of MCP Authorization 2026-07-28: an agent host discovers where a tasks server's authorization server is, identifies itself with a client ID metadata document, runs the OAuth 2.1 code flow with PKCE and a resource indicator, and presents the token the server was issued for. The server validates it and uses a token of its own for the API behind it.
User AlexMCP client Agent hostMCP server Tasks MCP serverAuthorization server auth.tasks.exampleResource Tasks API
An enterprise identity provider brokers the token
The same agent host and tasks server, with the company's identity provider deciding who may use which MCP server. The host signs Alex in at the identity provider, exchanges the ID token for an Identity Assertion JWT Authorization Grant (ID-JAG) addressed to the server's authorization server, and presents that grant for an access token. No consent screen at the tasks server: the policy lives at the identity provider.
User AlexMCP client Agent hostIdentity provider Enterprise IdP (idp.corp.example)Authorization server auth.tasks.exampleMCP server Tasks MCP server
An orchestrator delegates to a sub-agent with token exchange
Alex asks an orchestrator agent for a report. The orchestrator holds a token for the documents API in Alex's name and hands part of the work to a research sub-agent. Each hop exchanges the token it holds for a narrower one with an act claim, so the API can see the whole chain of who acted for whom, and no hop can do more than the one before it.
User AlexAgent Orchestrator agentAuthorization server auth.docs.exampleAgent Research sub-agentResource Documents API
How to read a chain
Actors are columns, steps are arrows. A step's citation is the section of the revision it follows; a step that carries a token shows it decoded, and a claim is marked added or changed when it differs from the token one hop earlier. A toggle replaces or removes steps; the page says which, and names the atlas requirements the toggle defeats so you can see how many listed servers meet them. Scenarios are illustrative of the specifications, not of any product: the domains are examples and no real token is involved.
Scenarios are data: one YAML file each, validated at build time against the shared model, open under CC BY 4.0 as one JSON document. To add or correct one, edit the file in therepository.