Overview

---

Built on the Model Context Protocol

MCP-I is an extension of the Model Context Protocol, providing delegated identity to AI agents, enforced through verifiable credentials and a tamper-evident audit chain. This allows agents to act on behalf of users while still operating under secure, accountable identity boundaries. It retains full interoperability with existing MCP-compatible agents and services, while enabling a higher trust model for agent actions.

---

Core Services

MCP-I is composed of modular components that align with decentralized identity and secure delegation:

• DID Service – Issues and resolves Decentralized Identifiers (DIDs)
• Credential Service – Issues Verifiable Credentials (e.g. delegation)
• Delegation Service – Verifies and chains authority credentials
• Edge Verifier – Enforces identity + delegation checks before request passes to backend
• Audit Service – Tracks identity, credential, and delegation events
• Core Service – Orchestrates events, registry, crypto utilities

---

Component Roles

Core Components

• Principal/User: The entity delegating authority to the agent
• Agent: Software component acting on behalf of the principal
• Service/Resource: System providing resources or tools via MCP
• Edge Proxy/Verifier: Validates agent identities and delegations

Infrastructure Components

• DID Resolver: Resolves decentralized identifiers to DID documents
• Revocation Service: Checks credential revocation status
• Reputation Service: Provides agent reputation information

Integration with Core MCP

MCP-I extends the Model Context Protocol with identity and delegation capabilities while maintaining full compatibility:

• Transport Layer: Reuses MCP transports (HTTP/SSE, stdio)
• Message Format: Extends MCP message format with identity headers
• Tools & Resources: Compatible with existing tool and resource definitions
• Client-Server Model: Maintains the same client-server architecture

Data Flow in MCP-I

Registration Flow

Request Verification Flow (Level 1)

Direct Service Authentication (Level 2+)

Trust Boundaries

MCP-I establishes clear trust boundaries between components:

Trust Model

Security Assumptions

• Agents are untrusted until proven otherwise through cryptographic verification
• DID resolution infrastructure is reliable and trustworthy
• Private keys are properly secured by their respective owners
• Edge Proxy is a trusted component for Level 1 implementations

---

Orchestration & Extensibility

All services can be deployed independently or managed centrally. The Core Service:

• Dynamically activates and deactivates services needed for identity, credential issuance, and delegation pipelines
• Maintains a live directory of all available MCP-I service endpoints and capabilities
• Hosts event bus for system-wide communication

You can plug in capabilities such as:

• DID Methods — Choose how your agents and users are identified. did:key offers simplicity, did:web supports domain-based trust, and did:ion anchors identifiers to public blockchains.
• Credential Schemas — Tailor what your credentials assert: permissions, roles, identity metadata, or custom business rules.
• Revocation Strategies — Use privacy-preserving Bitstring Status Lists or integrate your own backend revocation system.
• SDKs & Runtimes — Bring identity-native behaviors into frontends, mobile agents, or cloud services with supported SDKs like KYA-OS.

---

Next Steps

• Identity Layer (DIDs)
• Credential Models (VCs)
• Delegation Layer
• Verification Protocol Overview