Mooncite MCP
Mooncite lets agents search local Claude Code, Codex, Pi and ChatGPT history and verify each citation against the source file
- 69 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The server reads all local agent session logs and the ChatGPT export without changing any of them
- The installer registers the server in client settings and creates a launcher in `~/.local/bin`
Install
- Works in
Manual install
npx --yes github:WhenMoon-afk/claude-memory-mcp#v4.0.7 installRegisters the server in detected Pi, OMP, Codex and Claude Code clients. Then `"$HOME/.local/bin/mooncite" status`.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The server reads all local agent session logs and the ChatGPT export without changing any of them
- The installer registers the server in client settings and creates a launcher in `~/.local/bin`
This is third-party code. Review the repository files before installing.
What it does
Mooncite, formerly Claude Memory MCP, searches authorized Pi, OMP, Claude Code, Codex and ChatGPT history stored on your machine. The server hands the agent exactly three tools. `mooncite_recall` finds evidence in a bounded set of local data and, for a query of two or more terms with no strong match, adds local embedding search. `mooncite_inspect` checks a locator against the current bytes of the source file. `mooncite_status` reports coverage and health without conversation text. A verified status confirms that the citation matches the source and does not prove the text is true. The installer registers the server in available Pi, OMP, Codex and Claude Code clients, and ChatGPT serves only as an export source. Everything is read-only.
Who it is for. Developers who ask an agent to recall past decisions and want the citation backed by the original conversation.
Good fit when
- You need to recall why a library was chosen a month ago and see the quote from that chat
- The agent must not rely on a paraphrase until the citation is checked against the original
- History lives in several tools and you want one search across all of them
- A ChatGPT export should become a source for local search
- You need an index coverage report without exposing conversation text
Not a fit when
- You do not work on Linux: the project requires Linux with procfs
- You need memory that decides what to remember on its own: Mooncite only searches history
- Your Node.js is older than version 24 and cannot be upgraded
Example request
Call mooncite_recall with the phrase "message queue choice" and check the found citation with mooncite_inspectMCP
- Transport
- stdio
- Authentication
- not required
Limitations
Linux with procfs, Node.js 24 or newer, `npm` and access to GitHub are required. Packaged local embeddings exist only for Linux x64, and other Linux architectures keep lexical search. After install each client with an `exact` registration must be fully restarted. Citation verification shows only a match with the source file, and the fact stays for you to judge. The last commit dates from September 2026 and the old `claude-memory-mcp` address is kept.
How to disable. Removal and purge commands are described in the repository `docs/operations.md`. Remove the Mooncite server entry from each client configuration.
FAQ
How do I install Mooncite?
Run `npx --yes github:WhenMoon-afk/claude-memory-mcp#v4.0.7 install`, then check the launcher with `"$HOME/.local/bin/mooncite" status` and restart the clients.
Which tools does the server expose?
Three: `mooncite_recall` for search, `mooncite_inspect` for checking a citation against the file and `mooncite_status` for coverage and health.
What does the verified status mean?
The citation matches the current source file. It does not prove that what was said in the chat was right.
Where does the history come from?
From local logs of Pi, OMP, Claude Code and Codex and from a ChatGPT export. The server reads them and never changes them.
How is it different from claude-mem?
claude-mem compresses sessions into project memory itself. Mooncite invents nothing and returns citations checked against source files.
Discussion
No questions or reviews yet.
Related
MCP serversMemoMind MCP gives Claude Code project memory, with stored facts, events and conclusions visible in a local dashboard.
A CLI that indexes memory, rules and sessions of Claude Code, Codex, Cursor and Gemini and hands durable decisions to every agent
A memory plugin for Claude Code and Codex: team knowledge from wikis, tickets and chat with citations and stale-fact checks.
Membase keeps one shared memory for AI agents that Claude Code, Cursor and Codex read through MCP, an SDK or the membase command
Local memory for Claude Code, Codex and Claude Desktop: one Markdown folder and one MCP server shared by all.
Local memory for agents: stores conversations and project files verbatim and retrieves them with semantic search over MCP
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Mooncite (this card) | no | not required | 69 | |
| Claude-Mem | no | 116,457 | compresses sessions into project memory itself, while Mooncite returns citations and verifies them against the source file | |
| Episodic Memory | no | 499 | a plugin for searching old chats, while Mooncite adds citation checks and covers Pi, Codex and ChatGPT |
README in short
The README explains the purpose, shows a "human installs, agent recalls" diagram and gives a single install command. The first search and the three tools are described, with links below to the protocol, operations, architecture and security docs.