The Zero-Trust Identity Layer for MCP & Autonomous Agents
AIP is the open-source identity protocol for MCP and autonomous AI agents — establishing cryptographic identities, policy enforcement, and audit trails across heterogeneous agent systems.
Visit the GithubAgent Identity Protocol (AIP)
Open-source standard for auth, attestation, and governance of AI agents. The IAM layer for AI.
aip-go
The reference implementation of the Agent Identity Protocol — policy enforcement proxy for MCP. Sudo for AI Agents.
aip-playground
A minimal Python demo that shows the Agent Identity Protocol (AIP) in action — with both a CLI demo and a web UI.
Cryptographic identities for every AI agent, bound to verifiable credentials. AIP agents prove who they are before any tool call is made.
Fine-grained, policy-based access control on every MCP tool call. Deny-by-default with auditable allow rules — sudo for AI agents.
Immutable, structured audit trails across heterogeneous agent systems. Full visibility into what agents did, when, and on whose behalf.
As AI agents proliferate across enterprise environments, the need for a standardized identity layer has never been more urgent. Without AIP, agents operate under borrowed human credentials, making it impossible to distinguish agent actions from human ones. This creates audit gaps, policy violations, and security blind spots that traditional IAM tools were never designed to address.
AIP sits as a transparent proxy between AI agents and MCP servers. Every agent is issued a cryptographic identity credential at registration time. When an agent invokes an MCP tool, the AIP proxy intercepts the request, verifies the agent's identity, evaluates configured policy rules, and either permits or denies the operation, logging every decision to an immutable audit trail. This deny-by-default model mirrors the principle of least privilege: agents can only do what their policy explicitly allows, and every action is on the record.
AIP is language-agnostic and built for heterogeneous environments. Whether you are running a single Claude instance against a local MCP server or orchestrating hundreds of agents across distributed cloud infrastructure, AIP provides a consistent, interoperable identity plane; the IAM layer the Internet of Agents has been missing.
Explore the full AIP specification, reference implementations, and community resources to get started building with the Agent Identity Protocol.