Skip to content

Why MPL

The Problem

Regulated enterprises are blocking AI agent deployments because they cannot answer a fundamental question: "Can we prove the agent did what we said?"

Current agent protocols—MCP (Model Context Protocol) and A2A (Agent-to-Agent)—provide transport and tool invocation but leave critical gaps:

Gap Impact
No schema enforcement Messages lack contracts for what they mean
No quality guarantees No SLOs for agent behavior or output correctness
No audit trails No provenance or tamper detection for compliance
No policy controls No enforcement of organizational rules at the protocol level

These gaps force AI agent pilots into 12-18 week compliance approval cycles, effectively blocking production deployment.

What MPL Adds

MPL is a semantic overlay—not a replacement. It augments MCP and A2A with typed semantics while keeping your existing agent infrastructure intact.

graph TD
    subgraph "Without MPL"
        A1[Agent] -->|untyped JSON| B1[MCP/A2A Server]
        B1 -->|untyped response| A1
    end
    subgraph "With MPL"
        A2[Agent] -->|typed call| P[MPL Layer]
        P -->|validate + enrich| B2[MCP/A2A Server]
        B2 -->|response| P
        P -->|validated + QoM report| A2
    end

MPL introduces five capabilities:

  1. Semantic Types (STypes) — Versioned, schema-backed contracts for every message
  2. Quality of Meaning (QoM) — Measurable quality metrics with enforceable thresholds
  3. AI-ALPN Handshake — Capability negotiation before work begins
  4. Semantic Integrity — BLAKE3 hashing for tamper detection across hops
  5. Policy Engine — Rule-based enforcement of organizational constraints

Value by Stakeholder

For CTOs

MPL unblocks AI agent deployment by providing compliance teams the semantic guarantees they need. Your existing investments in MCP/A2A infrastructure remain intact—MPL overlays, not replaces.

For CISOs

Every MPL message carries a semantic hash, provenance metadata, and a QoM report. These map directly to SOX, GDPR, HIPAA, and EU AI Act requirements. See the Compliance Mapping for details.

For Architects

MPL deploys as a sidecar proxy requiring zero code changes. It works with existing MCP and A2A infrastructure, adding semantic contracts at the protocol level. See Integration Modes.

For Engineers

Install the CLI, point the proxy at your MCP server, and get immediate traffic visibility. Schema learning starts automatically. See the Quick Start.

Why an Overlay Wins

Teams adopt MPL faster than a greenfield protocol because it:

  • Works with existing transports — Reuses MCP WebSocket/HTTP sessions and A2A channels
  • Adds minimal friction — One handshake and a compact envelope
  • Provides immediate value — Typed payloads, QoM scores, and provenance without rewriting orchestrators
  • Supports incremental adoption — Start in transparent mode, graduate to enforcement

Challenges MPL Addresses

1. Meaning Is Implicit and Brittle

Payloads rarely ship with canonical schemas. Tooling depends on human convention instead of enforceable contracts. Silent breaking changes are discovered in production after multi-step workflows have mutated state.

MPL response: Versioned Semantic Types with registry-backed schemas and provenance hashes.

2. Capability Negotiation Is Ad Hoc

MCP clients learn about tool availability only after first failure. There's no telemetry for capability downgrades and no way to negotiate quality expectations.

MPL response: AI-ALPN negotiation that explicitly selects protocol version, STypes, tools, QoM profiles, and policies up front.

3. Quality Is Unmeasured

Teams rely on manual QA. There's no shared vocabulary for quality—schema fidelity, groundedness, and deterministic behavior are invisible to the protocol.

MPL response: Quality of Meaning profiles with measurable metrics and clear breach semantics.

4. Governance and Ecosystem Friction

Without shared registries, each integration reinvents schema definitions. Audit requirements remain unmet because there's no machine-verifiable quality evidence.

MPL response: A curated registry with namespace governance, semantic checksums, and auditable policy profiles.

5. Operational Complexity

Support teams face long MTTR because semantic mismatches masquerade as transport failures. Observability tools measure latency and uptime, not semantic correctness.

MPL response: Typed envelopes with structured telemetry, error taxonomies, and semantic signatures that route incidents to root cause.