agentidentityprotocol
Agent Identity Protocol - Zero-trust security layer for AI agents. Policy enforcement proxy for MCP with Human-in-the-Loop approval, DLP scanning, and audit logging.
claude mcp add --transport stdio openagentidentityprotocol-agentidentityprotocol docker run -i openagentidentityprotocol-agentidentityprotocol
How to use
Agent Identity Protocol (AIP) is a Go-based MCP server that provides a policy-enforced proxy layer for AI agents interacting with real tools and services. It focuses on authenticating agents, issuing and validating attestation tokens, and enforcing dynamic access policies at tool-call time. Use this server to gate tool access (e.g., Docker containers, databases, GitHub actions) behind a cryptographically signed identity so that you can audit, revoke, and Granularly authorize agent actions. The included design emphasizes Layer 1 identity (who the agent is) and Layer 2 enforcement (what it’s allowed to do) via a token (AAT) and a policy engine. Configure your environment so agents present AATs, then allow or deny tool calls accordingly through the AIP Proxy to real tools such as Docker, Postgres, or GitHub APIs. In operation, your MCP clients (Cursor, Claude, autonomous agents) will connect through the proxy, present their AATs, and receive an allow/deny decision with an audit trail for every action.
How to install
Prerequisites:
- Docker installed on the host (recommended for quick start)
- Git
- Optional: Go toolchain if you plan to build from source
Installation steps (Docker-based):
- Pull the image (if published): docker pull openagentidentityprotocol-agentidentityprotocol
- Run the container in detached mode (example): docker run -d --name aip-proxy -p 8080:8080 openagentidentityprotocol-agentidentityprotocol
- Configure environment variables as needed (e.g., AIP registry URL, revocation list, policy definitions). Example placeholders:
docker run -d --name aip-proxy -p 8080:8080 -e AIP_REGISTRY_URL=https://example.com/aip-registry
-e AIP_POLICY_PATH=/config/policy.yaml openagentidentityprotocol-agentidentityprotocol
Building from source (Go) (optional):
- Clone the repository: git clone https://github.com/openagentidentityprotocol/agent-identity-protocol.git
- Navigate to the project directory: cd agent-identity-protocol
- Build the binary: go build ./...
- Run the binary (adjust paths as needed): ./agent-identity-protocol
Prerequisites for building from source:
- Go 1.20+ (or compatible)
- Dependencies listed in go.mod
- Git
Note: Adjust commands and image names to match your actual registry and tagging strategy.
Additional notes
Tips and notes:
- Ensure your AIP Registry and Revocation List are reachable by the proxy at runtime to enforce correct policy and revocation checks.
- When testing, start with a permissive policy and gradually tighten controls to avoid accidental blockages.
- Common environment variables to consider: AIP_REGISTRY_URL (where tokens and revocation data are published), AIP_POLICY_PATH (path to YAML/JSON policy definitions), AIP_LOG_LEVEL (debug, info, warn, error).
- If you encounter authentication failures, verify that the agent’s AAT includes the correct issuer, audience, and expiry, and that the proxy trusts the issuer public key.
- For production, prefer the Docker deployment approach for easier scaling and isolation; ensure persistent storage for logs and policy configurations.
Related MCP Servers
mcp-agent
Build effective agents using Model Context Protocol and simple workflow patterns
mcp-agent
Lightweight, focused utilities to manage connections and execute MCP tools with minimal integration effort. Use it to directly call tools or build simple agents within your current architecture.
cursor-feedback-extension
Save your Cursor monthly quota! Unlimited AI interactions in one conversation via MCP feedback loop.
shellguard
MCP server that gives LLM agents read-only shell access over SSH
backlog
Help coding agents and developers to keep track of a project's backlog by storing tasks as markdown in git.
agentgate
Approval workflows for AI agents