Skip to content

MGP Documentation

Welcome to the Memory Governance Protocol documentation. MGP is an open protocol that standardizes how AI runtimes write, recall, govern, and audit persistent memory across heterogeneous backends.

New here? Jump straight to the Getting Started guide — you'll have a governed memory gateway running in under five minutes.

What Is MGP?

MGP defines a unified contract for governed memory without becoming a memory database itself. Your agent runtime talks to one protocol, and any compliant memory backend just works — whether it's in-memory, file-based, graph-oriented, relational, vector, or a managed service.

MGP is a peer protocol to MCP. MCP handles tools and resources; MGP handles persistent memory. They complement each other and can coexist in the same runtime.

Reading Paths

I want to build something quickly

  1. Getting Started — install, run the gateway, write & search your first memory
  2. Python SDK — MGPClient, AsyncMGPClient, and helpers
  3. Examples Overview — runnable end-to-end demos

I want to understand the protocol

  1. Project Overview — goals, boundaries, and architecture
  2. Architecture — layered view and request flow
  3. Protocol Reference — full operation semantics
  4. Schema Reference — all JSON Schemas explained
  5. MGP vs MCP — how the two protocols relate

I want to integrate or deploy

  1. Reference Implementation — the Python gateway
  2. Adapter Guide — build your own adapter
  3. Sidecar Integration — bridge an existing runtime
  4. Deployment Guide — production deployment patterns
  5. Security Baseline — security considerations

I want to verify compliance

  1. Conformance Profiles — Core, Lifecycle, Interop, ExternalService
  2. Compliance Suite — run the test suite
  3. Adapters Overview — adapter compatibility matrix

Full Site Map

Overview

Protocol

Implementations

Quality & Integration