LLM Wiki Memory MCP
Local memory for coding agents: sessions turn into a git-versioned Markdown wiki and come back through an MCP server.
- 129 ★ on the repository
- Last commit
Extended access
Why this level
- Hooks and the schedule read conversations and write notes to disk, and text goes to the selected LLM provider when knowledge is sorted.
- Notes containing secrets from conversations land in the wiki and in git when tracking is enabled.
Install
- Works in
Manual install
Set up llm-wiki-memory in this project: fetch https://raw.githubusercontent.com/ctxr-dev/llm-wiki-memory/main/AI-INSTALL-PROMPT.md and follow it EXACTLY (it covers fresh install and update; if already installed, the same file is local at @.llm-wiki-memory/src/AI-INSTALL-PROMPT.md).
First README option: a prompt for the agent that covers install and update.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Hooks and the schedule read conversations and write notes to disk, and text goes to the selected LLM provider when knowledge is sorted.
- Notes containing secrets from conversations land in the wiki and in git when tracking is enabled.
This is third-party code. Review the repository files before installing.
What it does
LLM Wiki Memory saves agent conversations, turns them into daily notes, knowledge and lessons, and returns the right context in later sessions. Everything sits on your machine as plain Markdown files in the `.llm-wiki-memory/` folder, with no vector database, Docker or cloud. Search runs on embeddings computed on the device. The agent calls MCP tools such as `search_memory` and `recall_lessons`. Claude Code also uses hooks: at session start the agent gets a short briefing of where you stopped, and at the end the conversation is distilled into notes.
Who it is for. Developers whose Claude Code, Cursor or Codex relearn the project every day and who want memory without extra infrastructure.
Good fit when
- Tomorrow the agent must remember an architectural decision made today.
- The agent hit the same mistake twice and the correction should become a lesson.
- An approved multi-step plan must outlive the session with its progress checkboxes.
- A team wants a shared project wiki that each member keeps in their own git.
Not a fit when
- The project runs on Node.js older than 22.18 and the runtime cannot be upgraded.
- Conversations contain material that must not be stored on disk even locally.
- The notes folder is meant to live on a cloud drive, where files may be half replicated.
Example request
Remember that this project writes migrations with Drizzle, and recall related lessons before the next task.MCP
- Transport
- stdio
- Authentication
- not required
Limitations
Node.js 22.18 or newer and git are required. The embedding model weighs about 219 MB and downloads on first run; `embed.backend: lexical` in `settings.yaml` skips it. Hook-based auto capture works only in Claude Code, while Cursor, Codex and Claude Desktop get the same MCP tools and follow the rules the installer sets up. The wiki calls the LLM provider from `.env` to sort knowledge: the claude, codex or cursor-agent CLI, or the Anthropic, OpenAI and compatible APIs. Cloud drives such as Dropbox or iCloud can scramble files mid-session.
How to disable. Remove the MCP entry and the installer hooks, and stop the hourly cron or launchd job. Delete the `.llm-wiki-memory/` folder separately once the memory is no longer needed.
FAQ
How does it differ from RAG?
Instead of a vector database the memory is a tree of Markdown files with git history. Embeddings only rank search results and are computed locally.
What is saved without my consent?
Daily notes and knowledge are written automatically. The agent saves self-improvement lessons only after an explicit yes, one lesson per approval.
How do I install it into a project?
Paste the README command with the `AI-INSTALL-PROMPT.md` address into the agent. By hand, clone the repository into `./.llm-wiki-memory/src` and run `bootstrap.sh`.
What does the hourly schedule do?
The `--schedule hourly` job warms embeddings gradually in the background. Search works without it, but cold notes are embedded during the first query.
Which agents work with the memory?
Auto-capture hooks exist only for Claude Code. Cursor, Codex, Claude Desktop and other MCP clients get the same tools and rules, and the consent for saving lessons applies everywhere.
Discussion
No questions or reviews yet.
Related
MCP serversLocal persistent memory for coding agents: hooks record every session, and Claude Code, Codex or Cursor recall past decisions through 54 MCP tools
ai-memory retains session knowledge in a Markdown wiki, exposes MCP retrieval and passes context between agents and machines.
Long-term agent memory in local Markdown files with a link graph, semantic search and MCP access
A long-term memory MCP server for AI agents: facts fade along a forgetting curve and related records are compressed into a summary
mem-agent-mcp: local memory for an agent
mem-agent-mcp
An MCP server that answers questions about your Obsidian-style notes through Dria's local mem-agent model.
Agent memory as a graph of linked facts: decisions and errors outlive the session, with no vector database or paid API calls
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| LLM Wiki Memory (this card) | no | not required | 129 | |
| ai-memory | no | API key | 15,226 | also keeps a Markdown wiki; LLM Wiki Memory adds Claude Code hooks, consent for each lesson and local embeddings |
| NeuralMemory | no | not required | 243 | keeps memory as a graph of links; LLM Wiki Memory holds a tree of readable files with git history |
README in short
The repository is built on the @ctxr/skill-llm-wiki engine and reports 1961 tests. Optional background consolidation merges near-duplicate notes and archives stale ones without deleting them for good.