Positioning NoticeThis page provides a high-level overview of the MPLP architecture model. It is not a normative protocol specification.For formal definitions, see docs.mplp.io.
Specification

Protocol Architecture

The MPLP specification describes a four-layer architecture model (L1–L4) that separates protocol semantics, coordination, execution, and integration concerns. See docs for formal definitions.

Protocol Topology

The 4-Layer Model

A layered separation of protocol semantics, governance, execution, and integration.

L1

Core Protocol

Semantic Foundation (Normative)

Defines stable protocol semantics and canonical data structures. L1 is the normative foundation of MPLP (schemas, invariants, and lifecycle primitives) and is independent of any execution environment.

Versioned & Governed · RFC Required

L2

Coordination

Rules Layer (Normative)

Defines coordination and governance semantics for multi-agent execution. L2 constrains how agents collaborate, confirm actions, record structured traces, and comply with protocol-level duties.

Composable Modules · Duty Enforced

L3

Execution Runtime

Behavioral Layer (Non-Prescriptive)

Defines how MPLP semantics are realized during execution. L3 covers runtime concerns such as AEL loops, VSL logic, and the Project Semantic Graph (PSG) to keep long-running behavior governable and replayable.

Implementation Agnostic · Contract Driven

L4

Integration Layer

Boundary Layer (Optional)

Defines integration boundaries between MPLP and external systems (models, tools, APIs, infrastructure). L4 ensures integrations do not leak external concerns into the core protocol, and remain governed objects.

External Boundaries · Governed I/O

Cross-Cutting

Cross-Cutting Duties

Protocol-level responsibilities that apply across L1–L3 (and constrain L4 integrations), ensuring observability, safety, and consistency throughout the execution lifecycle.

Coordination

Ensures structured coordination between agents, managing control flow, task ownership, and collaboration semantics.

Error Handling

Defines standardized error semantics and recovery patterns, ensuring predictable behavior across all protocol-conformant executions.

Event Bus

Defines event emission and subscription semantics, enabling decoupled communication and observability across protocol layers.

Learning Feedback

Defines how execution outcomes and feedback signals can be captured for protocol-level learning and policy refinement.

Observability

Defines mandatory traceability and observability requirements for all protocol-conformant executions.

Orchestration

Defines how Plans are executed, ensuring steps are processed in correct dependency order while preserving protocol invariants.

Performance

Defines protocol-level constraints on resource consumption, latency bounds, and execution efficiency.

Protocol Versioning

Defines how protocol versions evolve while preserving backward compatibility and execution safety.

Security

Enforces protocol-level access control and execution isolation, ensuring agents operate within defined permission boundaries.

State Sync

Defines state synchronization semantics across distributed nodes, ensuring protocol consistency in multi-node deployments.

Transaction

Defines atomicity guarantees for multi-step operations, providing rollback semantics in case of partial failure.
See protocol modules
Agent OS Protocol →Governed Stack →Evidence Chain →
Canonical References
Definition →Normative Spec (Docs) →Source of Truth (Repo) →

This website provides discovery and positioning content only. Normative requirements live in Docs; the ultimate source of truth is the Repository.

Next Steps
Read the Spec →View Source of Truth →See Verification Evidence →

Evidence is evaluated through documentation-defined scenarios and replayable artifacts. This website does not certify or endorse implementations.

Positioning Notice: This website provides discovery and positioning content only. For formal protocol definitions, see docs.mplp.io. Source of truth: GitHub Repository.