# ZengLiangYi/ChatCrystal [Health: Active]

**Category:** 🧠 Knowledge & Memory  
**Repository:** https://github.com/ZengLiangYi/ChatCrystal  
**GitHub Stars:** 61  
**npm Downloads (last month):** 860  
**Views:** 5  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/zengliangyi-chatcrystal

## Description
Local-first AI PKM memory server for coding conversations. Imports Claude Code, Cursor, Codex CLI, Trae, and GitHub Copilot chats into notes, semantic search, tag graphs, Markdown exports, and reusable MCP memory. npx -y chatcrystal mcp

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "chatcrystal": {
    "command": "npx",
    "args": ["-y","chatcrystal"]
  }
}
```

## Documentation

## What ZengLiangYi/ChatCrystal does

ZengLiangYi/ChatCrystal is a local-first personal knowledge management system for developers who work through problems with AI coding tools. It collects conversations from Claude Code, Cursor, Codex CLI, Trae, and GitHub Copilot, then organizes them into notes containing titles, summaries, conclusions, code snippets, and tags.

The resulting knowledge base supports semantic retrieval, relation-aware expansion of search results, a tag graph based on normalized tag co-occurrence, and Markdown export. Agents can use the MCP interface to recall prior experience and write new reusable knowledge back into the system. Local storage keeps the application and its data on the machine or self-hosted deployment, while configured LLM and embedding providers handle generation and retrieval-related processing.

## How it works

ChatCrystal scans local data directories for supported AI coding conversations and imports the records into its knowledge base. Imported conversations can be distilled into structured notes, either individually or in batches through the CLI. Embeddings enable semantic search, while tags and their relationships provide another way to navigate related knowledge.

The MCP entry point runs over standard input and output:

```bash
npx -y chatcrystal mcp
```

The same installation provides a CLI for checking status, importing conversations, searching, browsing notes, summarizing, viewing configuration, and controlling the background service. The web interface is available at `http://localhost:3721` when the server is started with `crystal serve -d`.

## Setup and configuration

Node.js 24 or newer is required. The npm-based setup can be started with:

```bash
npm install -g chatcrystal
crystal serve -d
crystal import
```

Open `http://localhost:3721` after starting the service. For MCP use, start `crystal mcp`, or use the `npx` command without a global installation.

An LLM provider is required for summarization, and an embedding provider is required for semantic search. These are configured separately in ChatCrystal settings. The project includes first-class OrcaRouter support for LLM generation; its OpenAI-compatible base URL is built in, so configuration uses an API key and a selected available model.

The desktop release targets Windows. Docker Compose is also available for self-hosting, storing application data in the `chatcrystal-data` volume mounted at `/data`.

## Tools and capabilities

- Import conversations from Claude Code, Cursor, Codex CLI, Trae, and GitHub Copilot.
- Convert conversations into structured notes with summaries, conclusions, snippets, and tags.
- Search the knowledge base semantically with embeddings.
- Expand search results using related knowledge relationships.
- Explore a tag knowledge graph based on tag co-occurrence.
- Export knowledge as Markdown.
- Let MCP-compatible agents recall and write reusable development experience.
- Inspect server status and database statistics through the CLI.

## Limitations and notes

Semantic search depends on a configured embedding provider, while summarization requires an LLM provider. A language model used for generation cannot be assumed to provide embeddings. The standard Docker deployment binds to all host interfaces by default; public access should use an HTTPS reverse proxy, and deployments exposed to multiple devices require an API token setup.

_Full upstream README: https://allmcps.com/mcp/zengliangyi-chatcrystal/readme_