Skip to content

The open, vendor-neutral protocol for governing the lifecycle of AI agent systems. Standardizes planning, coordination, execution, confirmation, and traceability across multi-agent runtimes and tools — enabling interoperable, observable, production-grade agent systems.

License

Notifications You must be signed in to change notification settings

Coregentis/MPLP-Protocol

MPLP — Multi-Agent Lifecycle Protocol

Repository Role: Source of Truth
Protocol Version: v1.0.0 — FROZEN
License: Apache-2.0

MPLP is a vendor-neutral lifecycle protocol for AI agent systems — defining how agents are created, operated, audited, and decommissioned.

Not a framework. Not a runtime. Not a platform.


Conceptual Positioning (POSIX Analogy, Informative)

MPLP uses POSIX as a conceptual analogy for protocol-level interface standardization:

  • POSIX "standard interface" → MPLP "schema-defined lifecycle interface" (Context / Plan / Confirm / Trace)
  • POSIX contracts → MPLP module contracts & kernel duties (invariants, failure semantics, expected signals)
  • POSIX conformance tests → MPLP Golden Flows & evidence-based verdicts (evidence-only evaluation)
  • POSIX portability → MPLP portability across substrates (models / frameworks / runtimes / vendors)
  • POSIX logging/debugging → MPLP Trace & Evidence Pack (replayable, auditable artifacts)

Boundary: This is an analogy for conceptual positioning only. MPLP is not POSIX, does not implement an OS or runtime, and does not make any POSIX compatibility or certification claim.

For the detailed mapping table, see the Documentation (Non-Normative): "POSIX Analogy — Conceptual Mapping" at docs.mplp.io.


This Repository Is the Source of Truth

This repository hosts the canonical source artifacts of MPLP.

It is the single authoritative location for:

  • Protocol schemas (frozen)
  • Reference SDK source packages
  • Golden Flow tests & cross-language validation
  • Documentation source (docs.mplp.io)
  • Governance constitution files
  • Conformance & interoperability definitions

This repository is not a marketing site and not a tutorial.


Three-Entry Model

MPLP uses a three-entry model. Each surface has a strict, non-overlapping role.

Surface Role Purpose
Website Protocol Legitimacy & Evaluation Entry Establish MPLP as a defined, governed, evaluable protocol
Documentation Specification & Reference Explain how the protocol works
Repository (this) Source of Truth Schemas, code, tests, governance

Choose the correct entry:

Note: MPLP Validation Lab (if present) is evidence-based evaluation only — it does not host execution and is not certification.

Anchor Closure

Each entry point has stable anchors for cross-referencing:

Website Anchors:

Documentation Anchors:

Repository Anchors:

Disambiguation: MPLP = Multi-Agent Lifecycle Protocol (not a license). MPLP is not POSIX (conceptual lens only).


4-Layer Architecture (L1–L4)

MPLP is defined as a four-layer protocol topology:

Layer Name Scope
L1 Core Protocol (Normative) Schemas, invariants, lifecycle primitives
L2 Coordination & Governance (Normative) Module semantics and kernel duties
L3 Execution Runtime (Non-prescriptive) Runtime realization (evidence-producing behaviors)
L4 Integration Layer (Optional boundary) Governed integration boundaries with external systems

Boundary: L4 is optional and defines integration boundaries; it does not imply hosted execution, vendor endorsement, or certification.


Repository Structure (High-Level)

This is a navigation overview. The authoritative boundaries are defined by the repo contents and governance documents.

.
├── schemas/                  # Frozen protocol schemas (v2)
│   └── v2/                   # 10 module schemas + invariants + events
├── packages/                 # Reference SDKs
│   ├── npm/                  # TypeScript packages (@mplp/*)
│   ├── pypi/                 # Python SDK (mplp-sdk)
│   └── sources/              # SDK source code
├── tests/                    # Protocol validation suites
│   ├── golden/               # Golden Flow tests (GF + profiles)
│   ├── cross-language/       # Cross-language validation
│   ├── runtime-compat/       # Runtime compatibility tests
│   └── schema-alignment/     # Schema alignment tests
├── docs/                     # Documentation source (docs.mplp.io)
├── governance/               # Constitution & governance files
│   └── 01-constitutional/    # CONST-001 to CONST-006
├── examples/                 # Usage examples (multi-language)
├── scripts/                  # Build & validation scripts
└── assets/                   # Diagrams & brand resources

Protocol Status

  • Protocol Version: MPLP v1.0.0
  • Specification State: FROZEN (no breaking changes permitted)
  • Schemas: Frozen under schemas/v2/
  • Golden Flows: Implemented as protocol evaluation scenarios (Core + profile flows)
  • Reference Implementations: Partial by design (non-normative)

Note: Absence of a reference SDK implementation does not imply a module is experimental. Protocol obligations are defined by schemas and specifications, not by SDK coverage.


Test Suites (Protocol Validation)

MPLP provides test suites for protocol validation and interoperability evidence:

Golden Flows (Evaluation Scenarios)

Golden Flows validate lifecycle invariants via replayable evidence (not via vendor endorsement).

  • Core Golden Flows: flow-01flow-05
  • Profile Flows: sa-flow-*, map-flow-* (when applicable)

Validation Suites

  • Cross-Language Tests: Validate consistent behavior across TypeScript, Python, Go
  • Runtime Compatibility: Ensure runtime integrations produce evaluable evidence
  • Schema Alignment: Verify schema consistency across modules and consumers

Interoperability Levels (IL1–IL3)

MPLP defines Interoperability Levels (not certification tiers) to describe evaluable evidence boundaries:

Level Name Scope
IL1 Schema Data interoperability (valid instances)
IL2 Governance Module semantics & lifecycle interoperability
IL3 Behavioral Runtime evidence & Golden Flow interoperability

Any claim of "MPLP interoperability" must map to these definitions and be backed by evidence.

Disambiguation: These interoperability levels (IL1–IL3) are intentionally not the same as the L1–L4 architecture layers.
L4 (Integration Layer) is an optional boundary layer and is not an interoperability "tier"; it scopes how external systems are integrated without leaking concerns into the core protocol.


Reference SDKs (Non-Normative)

SDKs are maintained as reference consumers of the protocol (schemas remain authoritative).

TypeScript Standard Library (@mplp/*)

Primary packages include:

Package Role
@mplp/sdk-ts Main developer entry
@mplp/core Protocol primitives & types
@mplp/schema Schema validators
@mplp/coordination Coordination & state-machine helpers
@mplp/modules L2 governance module helpers
@mplp/conformance Conformance kit & validation tooling
@mplp/runtime-minimal Minimal reference runtime (evidence producer)
@mplp/devtools CLI & debugging tools

For current published versions, refer to npm package listings and/or the repository's version governance records.

Python SDK

Package Role
mplp-sdk Main developer entry

For current published version, refer to PyPI.


What This Repository Does NOT Do

To maintain strict boundaries, this repository does not:

  • Provide marketing or positioning narratives (use the Website)
  • Replace the official documentation site (use Documentation)
  • Offer certification or compliance guarantees
  • Endorse vendors, runtimes, or agent frameworks
  • Act as a tutorial or learning platform

Documentation

The official documentation is built from /docs and published at:

👉 https://docs.mplp.io

Documentation includes:

  • Architecture (L1–L4)
  • All L2 modules (Context, Plan, Confirm, Trace, Role, Dialog, Collab, Extension, Core, Network)
  • Golden Flows (evaluation scenarios)
  • Standards mappings (ISO / NIST / W3C — informative only)
  • Governance, versioning, and change control
  • POSIX analogy mapping (informative / non-normative)

Governance

  • Protocol Governance: MPGC (MPLP Protocol Governance Committee)
  • Constitution Files: Located in governance/01-constitutional/
    • CONST-001 — Entry Model Specification
    • CONST-002 — Document Format Specification
    • CONST-003 — Frozen Header Specification
    • CONST-004 — Doc Audit Methodology
    • CONST-005 — Authoring Constitution
    • CONST-006 — Document Type Outlines & Depth Rules

Any change that affects schemas or protocol semantics follows the MPGC process.


Contributing

This repository accepts contributions within defined governance boundaries.

Before proposing changes, review the constitution files in governance/01-constitutional/.

Pull requests that introduce new protocol semantics, schema changes, or compliance/certification claims will be reviewed under MPGC governance.


License

Apache License 2.0
© 2026 Bangshi Beijing Network Technology Limited Company


Final Note

If you are reading this README expecting a tutorial or a sales pitch, you are intentionally in the wrong place.

This repository exists to ensure that:

MPLP remains observable, governed, and vendor-neutral at the protocol level.

About

The open, vendor-neutral protocol for governing the lifecycle of AI agent systems. Standardizes planning, coordination, execution, confirmation, and traceability across multi-agent runtimes and tools — enabling interoperable, observable, production-grade agent systems.

Topics

Resources

License

Code of conduct

Contributing

Security policy

Stars

Watchers

Forks

Packages

No packages published