OpenA2A AIM (Agent Identity Management) and OAuth 2.0
OAuth 2.0 is an authorization framework from the IETF, published as RFC 6749 in October 2012 and extended by later RFCs. It defines how a client obtains an access token, for a resource owner or on its own behalf. A person is not required: under the client credentials grant a program requests a token with its own credentials. AIM is an open source platform for AI agent identity. It keeps a registry of agents: each agent has an Ed25519 key, a trust score and a list of capabilities checked through an SDK, and MCP servers are attested by the agents that use them.
The two are not alternatives. AIM uses OAuth 2.0 where it issues tokens: AIM's agent token endpoint accepts the JWT-bearer grant type of RFC 7523 and returns a bearer token. The registry, the trust score, the action checks and the attestations are not defined by the RFCs cited on this page. AIM is not a general OAuth authorization server: it issues tokens for its own agent API only, and it implements neither the client credentials grant nor token exchange.
Checked on against the RFCs named in each row, as published by the RFC Editor and AIM at commit 3ab8285e, opens in a new tab. An RFC does not change once it is published, so each source names the RFC, the section and the month of publication. A later RFC can update an earlier one: RFC 9700 updates RFC 6749 and RFC 6750. OAuth 2.1 was an Internet-Draft (draft-ietf-oauth-v2-1-16) on the day of the check, not an RFC, so the rows describe OAuth 2.0. To report a line that is wrong, use the contact page.
Capability by capability
Signing people in
OAuth 2.0
Outside OAuth 2.0, which issues access tokens and does not sign people in. RFC 6749 says: "The way in which the authorization server authenticates the resource owner (e.g., username and password login, session cookies) is beyond the scope of this specification." OpenID Connect, a separate specification, adds sign-in: "OpenID Connect 1.0 is a simple identity layer on top of the OAuth 2.0 protocol."
Source: RFC 6749, section 3.1, October 2012, opens in a new tab, OpenID Connect Core 1.0 incorporating errata set 2, opens in a new tab
AIM
The dashboard has local email and password sign-in. No single sign-on, OpenID Connect or SAML login is wired in the open source build: the repository contains Google, Microsoft and Okta provider code, but no login route calls it.
Machine clients
OAuth 2.0
Defined. The abstract of RFC 6749 says the framework lets an application obtain access to an HTTP service "on behalf of a resource owner" or "on its own behalf". It defines a resource owner as "An entity capable of granting access to a protected resource. When the resource owner is a person, it is referred to as an end-user." Under the client credentials grant, "The client can request an access token using only its client credentials (or other supported means of authentication) when the client is requesting access to the protected resources under its control". RFC 7591 defines "mechanisms for dynamically registering OAuth 2.0 clients with authorization servers." None of the RFCs cited on this page uses the term AI agent: an AI agent that requests a token is a client, and the term client "does not imply any particular implementation characteristics (e.g., whether the application executes on a server, a desktop, or other devices)."
Source: RFC 6749, October 2012, opens in a new tab, RFC 6749, section 1.1, October 2012, opens in a new tab, RFC 6749, section 4.4, October 2012, opens in a new tab, RFC 7591, July 2015, opens in a new tab
AIM
Each agent is registered with its own Ed25519 public key. The token endpoint accepts one grant: a JWT-bearer assertion signed with that key, in exchange for a bearer token for AIM's own agent API. AIM does not implement the client credentials grant: the token endpoint refuses it.
Source: agent_service.go, opens in a new tab, oauth_token_handler.go, opens in a new tab
Client authentication with keys
OAuth 2.0
Defined. RFC 6749 says confidential clients "are typically issued (or establish) a set of client credentials used for authenticating with the authorization server (e.g., password, public/private key pair)." RFC 7523 defines a JWT as a client credential, sent as client_assertion with the type urn:ietf:params:oauth:client-assertion-type:jwt-bearer. RFC 8705 defines client authentication with mutual TLS, "based on either self-signed certificates or public key infrastructure (PKI)."
Source: RFC 6749, section 2.3, October 2012, opens in a new tab, RFC 7523, section 2.2, May 2015, opens in a new tab, RFC 8705, February 2020, opens in a new tab
AIM
An agent can sign each request with its Ed25519 key. On routes behind the request-signature check, the server verifies the signature over the method, path, timestamp and body, and rejects a timestamp more than 30 seconds from its own clock. The action-check request is signed over the action's fields (agent, capability, resource, context and timestamp) and verified against the agent's registered key. Agent routes also accept an API key or a bearer token in place of a signature.
Grant types
OAuth 2.0
RFC 6749 defines four grant types (authorization code, implicit, resource owner password credentials and client credentials) and a way to define more. RFC 7523 adds one in which the client presents a JWT as the grant, with the grant type urn:ietf:params:oauth:grant-type:jwt-bearer. RFC 9700, the security best current practice, restricts two of the original four: the resource owner password credentials grant "MUST NOT be used", and "clients SHOULD NOT use the implicit grant" unless conditions the RFC sets out are met.
Source: RFC 6749, section 1.3, October 2012, opens in a new tab, RFC 7523, section 2.1, May 2015, opens in a new tab, RFC 9700, section 2.4, January 2025, opens in a new tab, RFC 9700, section 2.1.2, January 2025, opens in a new tab
AIM
The agent token endpoint accepts one grant type, urn:ietf:params:oauth:grant-type:jwt-bearer, and refuses any other with unsupported_grant_type. It reads the JWT from client_assertion, the parameter RFC 7523 defines for client authentication, and not from assertion, the parameter RFC 7523 defines for this grant. The JWT must be signed with the agent's registered Ed25519 key, its sub must equal client_id, and it must carry an exp claim and an aud claim that names the AIM server. The response is a bearer token with expires_in set to 3600.
Source: oauth_token_handler.go, opens in a new tab, validateAssertionClaims, opens in a new tab
Token exchange and delegation
OAuth 2.0
Defined in RFC 8693, which specifies "how to request and obtain security tokens from OAuth 2.0 authorization servers, including security tokens employing impersonation and delegation." Its act claim "provides a means within a JWT to express that delegation has occurred and identify the acting party to whom authority has been delegated." A chain is written by nesting one act claim in another, and the nested claims "serve as a history trail that connects the initial request and subject through the various delegation steps undertaken before reaching the current actor."
Source: RFC 8693, January 2020, opens in a new tab, RFC 8693, section 4.1, January 2020, opens in a new tab
AIM
OAuth token exchange (RFC 8693) is not implemented: the agent token endpoint accepts the JWT-bearer grant only. Delegation exists outside OAuth, in the TypeScript SDK, which creates and verifies Ed25519-signed delegation chains between keys. The AIM server does not issue or check them.
Source: oauth_token_handler.go, opens in a new tab, delegation.ts, opens in a new tab
Proof of possession
OAuth 2.0
Defined, in extensions. RFC 6750 defines the bearer token, which any holder can use: "Using a bearer token does not require a bearer to prove possession of cryptographic key material (proof-of-possession)." RFC 9449 (DPoP) describes "a mechanism for sender-constraining OAuth 2.0 tokens via a proof-of-possession mechanism on the application level", and RFC 8705 binds an access token to the client's mutual-TLS certificate. RFC 9700 says "Authorization and resource servers SHOULD use mechanisms for sender-constraining access tokens".
Source: RFC 6750, section 1.2, October 2012, opens in a new tab, RFC 9449, September 2023, opens in a new tab, RFC 8705, February 2020, opens in a new tab, RFC 9700, section 2.2.1, January 2025, opens in a new tab
AIM
DPoP is not found in the AIM repository at this commit. Tokens from the token endpoint are plain bearer tokens. Requests authenticated by signature are signed per call, as in the row on client authentication.
Permissions
OAuth 2.0
Defined as scope, a list of strings the client sends with its request: "The strings are defined by the authorization server." RFC 9396 adds authorization_details, a parameter "used to carry fine-grained authorization data in OAuth messages", and RFC 8707 adds a resource parameter that names the service or resource a token is requested for.
Source: RFC 6749, section 3.3, October 2012, opens in a new tab, RFC 9396, May 2023, opens in a new tab, RFC 8707, section 2, February 2020, opens in a new tab
AIM
Each agent has a list of granted capabilities. The SDK, running inside the agent process, sends each wrapped action to the server, which compares the capability name with that list. In the default mode (monitoring) a capability outside the list is allowed and, under the default policy, raises an alert. The Python SDK's decorator also registers a capability on first use, and in this mode the server grants it automatically. In strict mode a capability outside the list is denied under the default policy, and the SDK blocks the call.
Source: agent_service.go, opens in a new tab, enforcement.py, opens in a new tab
Registry of agents with a trust score
OAuth 2.0
Not found in the RFCs cited on this page, checked 2026-10-12: none defines a risk or trust score for a client. Nearest defined mechanisms: client registration, in the row on machine clients, and the authorization server's own policy, under which it "MAY fully or partially ignore the scope requested by the client, based on the authorization server policy or the resource owner's instructions."
Source: RFC 7591, July 2015, opens in a new tab, RFC 6749, section 3.3, October 2012, opens in a new tab
AIM
Each agent has a trust score calculated from nine weighted factors; a factor with no data is left out of the weighted sum. It is recalculated on events such as registration, verification by an administrator, capability changes and capability reports from the SDK. A capability violation also lowers the stored score directly. Each action check reads the stored score.
MCP servers
OAuth 2.0
MCP is not mentioned in the RFCs cited on this page, the latest of which is from January 2025. The MCP specification builds its authorization on OAuth. In revision 2026-07-28, "A protected MCP server acts as an OAuth 2.1 resource server" and "Authorization servers MUST implement OAuth 2.1 with appropriate security measures for both confidential and public clients." Authorization is optional for an MCP implementation, and the documents that page lists include RFC 6750, RFC 7591 and RFC 8707.
Source: MCP specification 2026-07-28, Authorization, opens in a new tab
AIM
AIM is not an OAuth authorization server for MCP. Agents submit signed attestations of the MCP servers they connect to, and the server checks each signature against the agent's registered key. A drift alert is raised when a capability detection run finds that a server's manifest has changed since the last run.
Source: mcp_attestation_service.go, opens in a new tab, mcp_manifest_service.go, opens in a new tab
Workload identity
OAuth 2.0
Not found in the RFCs cited on this page, checked 2026-10-12: none names SPIFFE or a Kubernetes service account token. Nearest defined mechanism: under RFC 7521 a client can present an assertion it obtains "from a third-party entity capable of issuing, renewing, transforming, and validating security tokens", and RFC 7523 defines the JWT form of that assertion, as a grant and for client authentication.
Source: RFC 7521, section 3, May 2015, opens in a new tab, RFC 7523, section 2.1, May 2015, opens in a new tab, RFC 7523, section 2.2, May 2015, opens in a new tab
AIM
Not implemented for agents at this commit: AIM accepts no SPIFFE identity, Kubernetes service account token or external OpenID Connect issuer as an agent credential.
Source: oauth_token_handler.go, opens in a new tab, pqc_agent_auth.go, opens in a new tab
Audit
OAuth 2.0
Not found in the RFCs cited on this page, checked 2026-10-12: none defines an audit log. Nearest defined mechanisms: the nested act claims of RFC 8693, in the row on token exchange, and token introspection in RFC 7662, by which a protected resource asks an authorization server "to determine the active state of an OAuth 2.0 token and to determine meta-information about this token."
Source: RFC 8693, section 4.1, January 2020, opens in a new tab, RFC 7662, October 2015, opens in a new tab
AIM
An action check that passes its signature check is written to the audit log as a row with the agent, the capability, the resource, the stored trust score and whether it was allowed. The write is best effort: if it fails, the check still returns its decision. The log is not append-only at the database layer and is not signed.
Source: verification_handler.go, opens in a new tab, audit_log_repository.go, opens in a new tab, SECURITY.md, opens in a new tab
License and hosting
OAuth 2.0
A set of specifications, not software: there is no server to install or host. RFC 6749 is an Internet Standards Track document and RFC 9700 is a Best Current Practice. Both come from the Internet Engineering Task Force (IETF). The authorization servers, client libraries and resource servers that implement them are separate software.
Source: RFC 6749, October 2012, opens in a new tab, RFC 9700, January 2025, opens in a new tab
AIM
The repository is Apache-2.0. Self-hosted with Docker Compose. Also offered as a hosted service, AIM Cloud, which is built from a private mirror of the repository with cloud-only additions.
Source: LICENSE, opens in a new tab, docs/DEPLOYMENT.md, opens in a new tab, docker-compose.quickstart.yml, opens in a new tab
Which layer each one covers
OAuth 2.0
- Issuing access tokens to clients, for a resource owner or for the client itself.
- Authenticating clients, with a secret, a JWT or a TLS certificate.
- Limiting a token by scope, by resource and by the key or certificate of its holder.
- Passing authority from one party to the next with token exchange.
AIM
- Keeping a registry of agents, each with its own key.
- Scoring each agent and recording the action checks it submits.
- Checking each action an agent submits against its granted capabilities.
- Recording which MCP servers agents attest to, and alerting when a detection run finds a server's manifest changed.
A team can run both: an OAuth 2.0 authorization server for the access tokens its APIs and MCP servers accept, AIM for its registry of agents and their action checks. AIM has no integration with an external authorization server at this commit: it accepts no token from an external OpenID Connect issuer as an agent credential, and no single sign-on is wired in to its dashboard in the open source build. Agents authenticate to AIM with their own keys or an AIM API key, and in the open source build people sign in to the AIM dashboard with a local account.
Registering an agent with AIM
In the Python SDK, secure() registers an agent and perform_action wraps the actions to be checked. secure() needs credentials first, from aim-sdk login or an SDK downloaded from the dashboard. For a self-hosted server, AIM's README at this commit says to create the agent in the dashboard first. The capability name on the decorator is what the server compares with the agent's grants. The resource is recorded with the check and is not compared.
from aim_sdk import secure
agent = secure("my-agent")
@agent.perform_action("db:read", resource="users_table")
def get_user_data(user_id):
...