GitHub radar
Using MCP to Store Persistent Agent Memory for Claude Code
OKF Agent Memory is a Git-native MCP server that provides AI coding agents with persistent memory via the Open Knowledge Format (OKF) v0.2, storing project context as Markdown files in your repo using 80% fewer tokens than before.
According to the developers, OKF Agent Memory is ‘a Domain-Neutral, Git-Native Persistent Project Memory for AI Agents based on the Open Knowledge Format (OKF) v0.2’. The Open Knowledge Format is a memory format created by Google. Without persistent memory, an agent forgets all architectural details and session state when a context window expires. OKF Agent Memory saves project memory as simple Markdown files with YAML front matter in a ‘knowledge/’ directory in your repo which get loaded automatically when you start a new session. It embeds an MCP server so you can run Claude Code or any other MCP compatible agent without the need for external database servers or services. It implements a Progressive Disclosure mechanism so that only relevant context information gets presented to the agent at any point in time. According to the developers, this results in an 80% reduction in token usage for long projects. It has an in-memory BM25 search index with a sub-300 microsecond response time. It works with local and cloud-based LLMs like LM Studio and Ollama.
Why a vibe-coder should care
Every time the context window resets, the agent can resume work from where it left off without needing to explain the project layout, architecture choices or conventions. Long sessions are also less expensive since the agent can retrieve specific pieces of information from memory rather than having to read through the entire codebase.
How to install
Copy this and send it to your agent — Claude Code, Codex, any of them:
Set up the OKF agent memory MCP server from https://github.com/okf-memory/okf-agent-memory following the README, initialize a knowledge/ folder in the current project, connect to Claude Code, and show me a test memory entry with a search result — ask me for any permissions you need.
Runs on a regular laptop — just Go or a prebuilt binary, no external services required.
Open on GitHub▌ More finds