Skip to main content

Topic

AI agent identity & MCP authorization

AI agents need identities of their own, but they may also act on behalf of users and organizations. Learn how to identify and authenticate agents, manage their lifecycle, delegate authority, handle tokens and consent, and control access across MCP clients, servers, and tools while maintaining human oversight and a clear audit trail.

Articles

6 of 8 published

  1. How authorization works across an MCP client, server, and tool call, and the OAuth 2.1 flow the spec builds on.

  2. What OAuth 2.1 tightens up compared to OAuth 2.0, and exactly what an MCP server or client has to implement.

  3. A step-by-step walkthrough of discovery, dynamic registration, and the authorization code flow for an MCP client.

  4. A framework-specific walkthrough of wiring LangGraph’s MCP client up to an OAuth 2.1-protected server.

  5. Why a server deciding what to trust and a client proving who it acts for are separate problems with separate fixes.

  6. Per-user credentials in an MCP tool call

    Why an MCP server usually can’t reuse the client’s token, and how token exchange scopes a call to one specific user.

    Coming soon
  7. How a short-lived access token breaks a long agent run partway through, and how to design around it.

  8. MCP auth options compared

    API keys, OAuth 2.1, and mTLS for MCP servers, compared on security, setup effort, and interoperability.

    Coming soon

Secure agentic identity with Choreo

Choreo Identity and Access Manager extends OAuth 2.0 and delegated authorization to non-human identities, including AI agents and MCP-based workflows.

Explore Choreo IAM