Integrate Mneme

Choose the interface, then wire Mneme into your stack

Start with MCP for agent interoperability, the CLI for repository and CI enforcement, or the Python package for custom application workflows. Then use the integration matrix for the boundary your coding system actually exposes.

Three supported entry surfaces

This is the canonical starting point for integrating Mneme. The interfaces share the same decision model, but they serve different jobs and do not imply the same authority or enforcement boundary.

Agent interoperability

Model Context Protocol

Use Mneme's local stdio Decision MCP when an MCP-capable host needs to search decisions, inspect applicability context, trace lineage, or submit non-authoritative proposals. MCP does not accept or activate decisions.

Best for: MCP hosts, coding agents, decision exchange

MCP reference →Integration guide →

Repository control

CLI

Use mneme when a repository, hook, developer workflow, or CI pipeline needs deterministic checks, audits, protection activation, proposal review, or integration commands.

Best for: local workflows, hooks, CI gates, operators

CLI reference →Protection workflow →

Embedded workflows

Python package API

Use the installed mneme package to compose Mneme inside custom Python adapters and agent workflows. Core package surfaces include decision storage, retrieval, context construction and pipeline components; shipped adapters show the supported integration patterns.

Best for: custom adapters, Python agent frameworks, application embedding

Python package source →SDK integration example →

Interoperability — decision exchange

A protocol boundary for retrieving decisions and submitting non-authoritative proposals. Connection is not enforcement, and MCP does not receive human decision authority.

Local stdio · Shipped in 0.9.0

Model Context Protocol

Six frozen Decision Index tools for retrieval and proposal submission, with accept, reject, activation, and supersession deliberately excluded.

Enforcement — native and validated

A shipped adapter checks proposed changes against the decision corpus before they reach disk. Maintained code, tests on main, published evidence.

Native · Shipped

Claude Code

PreToolUse hook blocks violating Edit, Write, and MultiEdit calls before they reach disk.

Native · Shipped

Claude Agent SDK

Injects relevant decisions when a task starts; checks proposed mutations before they land.

Native · Shipped

Google Antigravity

Pre-tool adapter gates Antigravity’s file-mutation tools through the same engine. Tested on IDE 2.8.1.

Validated compatibility

Paperclip

Unmodified Mneme enforced under both Paperclip transports — CLI and ACP. No adapter required.

Native · Shipped

Codex CLI

Pre-execution enforcement for native apply_patch mutations, with a Stop whole-file audit covering shell and script-driven writes. Validated 10/10 against Codex CLI 0.149.1.

Native · Shipped

LangChain / LangGraph

Inject architectural context before model calls and deterministically govern supported write_file and edit_file mutations through LangChain agent middleware.

Native · Shipped

Kiro

Kiro CLI 3.0 / v3 native write and append mutations are governed before they reach disk.

Propagation — rules export and CI gates

The same corpus carried into tools without a runtime adapter: generated editor rules, and a deterministic gate in pipelines you own.

Rules export

Cursor

Generate conflict-resolved Cursor Rules from the corpus with one command.

CI gate

GitHub Actions

Run mneme check against every PR diff with a reference workflow you own.

CI gate

GitLab CI

The same deterministic gate, native to merge requests.

Corpus workflow

ADR Import

Bring your existing ADRs into the enforcement corpus in one command.

An engineer drawing a connected workflow on a whiteboard — **One corpus, many enforcement boundaries.** The workflow changes by tool; the architectural decisions do not.

Host environments

No dedicated adapters. Governance reaches these tools through the layers above — an underlying supported agent, generated rules, or CI.

Host environment

VS Code

Hosts Claude Code sessions; governed by the PreToolUse hook, no extension needed.

Host environment

GitHub Copilot

No hook API. Covered by the CI gate and the rules export.

Host environment

JetBrains IDEs

Covered by the corpus and the CI gate, regardless of IDE.

Host environment

Warp

No adapter. Governance applies via underlying agent hooks or a CI gate on the diff.

Experimental and planned

Nothing here carries production support yet. Experimental means code exists and specific boundaries are tested, but validation or coverage remains incomplete; planned means design work only.

Experimental

OpenCode

Plugin-hooks approach under evaluation; no production plugin. Compaction experiment: NULL verdict.

Experimental

Hermes Agent

Context injection and pre-tool enforcement for supported mutations. Experimental against Hermes Agent 0.19.0; shell/process bypasses remain and no blocking Stop-equivalent is available.

Middleware POC planned

Deep Agents

Mneme as custom middleware around filesystem and tool execution: pre-mutation blocking, change reconstruction, subagent propagation.

Planned

n8n

Governance checks for AI-driven workflow automation.

Works alongside — Perplexity Enterprise turns research rationale into enforceable decisions · Microsoft Agent Forge pairs an autonomous workflow substrate with deterministic governance. Research and ecosystem tools, not Mneme integrations.

Every layer reads the same file: .mneme/project_memory.json. Install with pip install mneme-hq>=0.9.2, scaffold with mneme init, and wire the layer that fits your stack. The architectural argument: governance across heterogeneous AI coding agents. Standards alignment: MCP, AGENTS.md, NIST CAISI.