Skip to content
1Claw Academy
← Curriculum
Intermediate

Agents & Access Control

Give agents an identity, scope them tightly, and connect over MCP.

By the end of this track

You’ll be able to give an agent a tightly scoped identity and let it call external APIs without ever holding a credential.

15 lessons~37 min49 questions
Start → Agents as first-class principals
  1. 1Agents as first-class principalsRegister an agent, get its automatic keypairs, and exchange its API key for a short-lived JWT.
  2. 2Policies and scoped accessGrant an agent read access to one secret path, then use deny rules and priority to carve exceptions out of a broad grant.
  3. 3Lab: find out what your policy actually matchesLabWrite three patterns you believe are narrow, then list what each one really reaches.
  4. 4Cedar and OPA policy backendsExpress authorization as declarative Cedar or Rego policy, and dry-run it before it ever denies a real request.
  5. 5Agent self-enrollment and approvalsLet an agent enroll itself with no credentials and route the approval to a human.
  6. 6MCP server integrationConnect an AI client to the 1Claw MCP server and run a first list_secrets call.
  7. 7Execution Intents and bindingsLet an agent call an external API through a binding, so the credential is injected server-side and never enters the model's context.
  8. 8Lab: build a credential brokerLabWrite both patterns side by side, dump each agent's context, and see the difference the broker makes.
  9. 9Agent-to-human sharingLet an agent hand a credential it discovered or generated back to the human who created it, without knowing their email.
  10. 10Peer memory: access control for what nobody grantedSeveral agents can share a model of one human. That model is derived, not granted, which changes what access control has to mean.
  11. 11Sub-agents and delegationLet one agent hand work to another under a human-authored delegation, with tool allowlists, depth limits, and daily caps.
  12. 12Lab: watch a delegation chain run awayLabLet agents delegate onward and see what the depth limit is actually stopping.
  13. 13Agent discovery and A2APublish an agent card to the public directory so other agents can find it over A2A and MCP.
  14. 14Lab: fetch a real agent's public cardLabPull a live agent's public discovery card and run the same disclosure check the discovery lesson described, against something you did not write.
  15. 15Capstone: design an access modelCapstoneTake a real brief and produce the identities, policies, and bindings a working system needs.