State Drift
Agents lose intent and constraints as state transitions occur without formal invariants.
Protocol Specification v1.0.0
Multi-Agent Lifecycle Protocol
The Agent OS Protocol
The Lifecycle Protocol of AI Agents
The lifecycle protocol for AI agent systems.
MPLP defines how agents are created, operated, audited, and decommissioned across their full lifecycle.
Not a framework. Not a runtime. Not a platform.
Status
Frozen / Stable
License
Apache 2.0
Governance
MPGC Managed
Protocol Status: Frozenv1.0.0-stable
The Governance Gap
The bottleneck is not intelligence. It is lifecycle governance.
Multi-agent systems fail not because agents are weak, but because lifecycle semantics are undefined. These failures are structural outcomes of missing protocol-level invariants.
Frameworks scale features.
Protocols scale ecosystems.
Hallucination Accumulation
Errors compound across agent boundaries due to missing semantic validation frames.
Orchestration Collapse
Coordination complexity grows exponentially without a unified lifecycle protocol.
Audit Black Holes
Traceability is lost when lifecycle events are not governed by a canonical standard.
Protocol Topology
A Protocol Stack, Not a Framework Stack
MPLP sits above agent frameworks and below applications, defining the normative lifecycle semantics that every conformant system must respect.
Core Protocol
Lifecycle primitives and semantic invariants.
Coordination
Governance primitives: Context, Plan, Confirm, Trace.
Execution
AEL loops, VSL logic, and Project Semantic Graph.
Integration
Models, tools, and external system adapters.
Adopt MPLP Incrementally
Modules are not features. They are lifecycle constraints.
MPLP is designed for gradual, low-risk adoption. Each module is not a capability — it is a lifecycle constraint. Partial adoption is safe by design.
Start with Trace
Capture structured lifecycle events and decision history.
Add Confirm
Gate critical actions with explicit confirmation and auditability.
Introduce Plan & Context
Enforce structured intent, plans, and lifecycle consistency.
Implementation
Governing an agent takes minutes, not months.
MPLP is model-agnostic and framework-neutral. You can start by governing a single agent state and expand as your system grows.
01
Install Protocol SDK
npm install @mplp/sdk-ts
02
Record Lifecycle Event
import { Trace } from "@mplp/sdk-ts";
Trace.record({
event: "intent.created",
detail: { description: "Generate report" }
});Protocol Conformance
Golden Flows: The Interoperability Standard
Golden Flows are not examples — they are the normative conformance tests of the MPLP protocol. They ensure cross-vendor interoperability and semantic consistency.
Ecosystem Topology
Built for Architects, Developers, and Auditors
MPLP provides the structural foundation for building observable and auditable agent systems. Canonical SDKs and schemas ensure rapid, safe integration.
Governance
Evidence Chain
Plan → Confirm → Trace: audit-ready evidence for agent systems, aligned with ISO/IEC 42001 and NIST AI RMF.
Explore Governance
Resources
SDKs & Schemas
Canonical SDKs and module schemas to implement MPLP incrementally without vendor lock-in.
Explore Ecosystem
Evolution
Versioning Policy
Transparent evolution policy for the frozen v1.0.0 specification and forward-compatible extensions.
View Governance
Implement the Standard.
Build agent systems that remain reliable, observable, and governable — even as models, frameworks, and vendors change.
Protocol v1.0.0→Apache 2.0→MPGC Governed