Core Concepts

MPL is built on a small set of interlocking concepts. Each concept addresses a specific challenge in governing AI agent communication within regulated environments. Together, they form a complete semantic governance layer.

---

Concept Map

The following diagram shows how MPL's core concepts relate to each other:

graph LR
    STypes[STypes<br/>Semantic Types] -->|registered in| Registry[Registry<br/>Schema Store]
    STypes -->|carried by| Envelope[Envelope<br/>Message Wrapper]
    Envelope -->|evaluated by| QoM[QoM<br/>Quality Metrics]
    QoM -->|enforced by| Policy[Policy Engine<br/>Rule Enforcement]
    Registry -->|negotiated via| Handshake[AI-ALPN<br/>Capability Negotiation]
    Handshake -->|selects| QoM
    Policy -->|references| Registry
    Envelope -->|hashed for| Audit[Audit Trail<br/>Provenance]

Reading the Concept Map

Start from STypes on the left. They define what a message means. The Registry stores their schemas. The Envelope wraps messages with SType metadata. AI-ALPN negotiates which STypes and profiles both parties support. QoM measures semantic quality. The Policy Engine enforces organizational rules based on all of the above.

---

Concepts at a Glance

Concept	Description	Learn More
Architecture	The three-layer protocol stack and eight architectural pillars that define MPL's design	Architecture
Semantic Types (STypes)	Globally unique, versioned identifiers backed by JSON Schema that define message semantics	STypes
Quality of Meaning (QoM)	Six measurable metrics with configurable profiles that quantify semantic quality	QoM
Integration Modes	Three deployment models (Sidecar, SDK, Native) for adopting MPL in any environment	Integration Modes
Envelope	The message wrapper carrying payload, SType, semantic hash, provenance, and QoM report	Architecture: Envelope
AI-ALPN	Capability negotiation handshake that aligns peers before work begins	Architecture: AI-ALPN
Policy Engine	Rule-based enforcement layer that gates actions on QoM thresholds, SType constraints, and org policies	Architecture: Policy
Registry	Versioned store of SType schemas, QoM profiles, and assertion libraries	Architecture: Registry

---

How the Concepts Fit Together

A typical MPL interaction exercises all concepts in sequence:

sequenceDiagram
    participant Agent
    participant MPL as MPL Layer
    participant Registry
    participant Server as MCP/A2A Server

    Note over Agent,MPL: 1. AI-ALPN Handshake
    Agent->>MPL: Propose STypes + QoM profiles
    MPL->>Registry: Resolve schemas
    MPL-->>Agent: Negotiated capabilities

    Note over Agent,MPL: 2. Typed Call via Envelope
    Agent->>MPL: Envelope(stype, payload, profile)

    Note over MPL: 3. Validate (SType schema)
    Note over MPL: 4. Evaluate (QoM metrics)
    Note over MPL: 5. Enforce (Policy rules)

    MPL->>Server: Forward validated request
    Server-->>MPL: Response
    MPL-->>Agent: Envelope + QoM Report

---

Where to Start

Recommended Reading Order

1. Architecture -- Understand the protocol stack and design pillars
2. Semantic Types -- Learn how messages get their meaning
3. Quality of Meaning -- See how quality is measured and enforced
4. Integration Modes -- Choose how to deploy MPL in your environment

---

Design Principles

MPL's concepts are guided by these principles:

Principle	Meaning
Transport Independence	Works with MCP, A2A, or any future protocol
Progressive Adoption	Start observing, then learn, then enforce
Zero Trust Semantics	Every message is validated regardless of source
Measurable Quality	Quality is numeric, not binary pass/fail
Auditability	Every decision leaves a cryptographic trace