# kunwar-shah/claudex [Health: Active]

**Category:** 🧠 Knowledge & Memory  
**Repository:** https://github.com/kunwar-shah/claudex  
**GitHub Stars:** 97  
**npm Downloads (last month):** 351  
**Views:** 2  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/kunwar-shah-claudex

## Description
Persistent memory + FTS5 full-text search for Claude Code conversation history. Indexes /.claude/projects/ JSONL into SQLite, exposes 10 MCP tools (store/recall/search memories, list sessions, get summaries) plus prompts. Includes a web UI for visual exploration with themes and exports. Works with Claude Code, Cursor, Codex, Windsurf, and any MCP-compatible client. npm install -g @kunwarshah/claudex

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

```json
"mcpServers": {
  "claudex": {
    "command": "npx",
    "args": ["-y","@kunwarshah/claudex"],
    "env": {
      "PORT": "",
      "PROJECT_ROOT": ""
    }
  }
}
```

**Requires environment variables:** `PORT`, `PROJECT_ROOT` — the values above are empty placeholders; fill in real credentials before running (see the repository for what each one is for).

## Documentation

## What kunwar-shah/claudex MCP server does

The kunwar-shah/claudex MCP server gives MCP clients access to Claude Code history and a persistent memory layer. It discovers conversation files beneath the Claude projects directory, reads their JSONL content, and makes that information available through MCP tools. The project also provides a browser-based interface for reviewing and analyzing the same conversation history.

Its memory model is intended for coding knowledge such as conventions, architectural decisions, recurring error patterns, and other project context. Stored entries can include a priority from 1 to 10, a confidence value, and a time-to-live so temporary information can expire.

## How it works

Claudex uses SQLite FTS5 for full-text indexing. Search can be narrowed by project, session, role, and date range, with matching content highlighted. The parser supports multiple Claude Code conversation formats, including V1, mixed V2, and V2.0 or later templates, with automatic format detection.

The MCP release exposes 10 tools covering project context, session search, conversation retrieval, and structured memory CRUD. Three prompts—`/recall`, `/catchup`, and `/history`—provide predefined ways to access earlier context. Responses support minimal, standard, and full detail levels for managing the amount of returned context.

## Setup and configuration

Install the package globally with npm:

```bash
npm install -g @kunwarshah/claudex
```

Register the kunwar-shah/claudex MCP server with Claude Code using stdio transport:

```bash
claude mcp add --transport stdio claudex -- claudex-mcp
```

The expected prerequisites are Node.js 18 or later, npm, and Claude Code conversation history in `~/.claude/projects`. `PROJECT_ROOT` changes the Claude projects directory, while `PORT` changes the web application port from its default of 3400. The CLI also accepts `--project-root` and `--port` options. An npx invocation is available when a global installation is not preferred.

For the web application, the repository can be cloned and started from source. Its backend listens on port 3400 by default, and the development frontend is available at `http://localhost:3000`.

## Tools and capabilities

The kunwar-shah/claudex MCP server supports:

- Project context lookup and session listing.
- Full-text search across indexed conversations.
- Conversation retrieval and session summaries.
- Structured memory creation, lookup, updating, and deletion.
- Context-size control through three detail levels.
- Three prompts for recalling, catching up on, and reviewing history.

The accompanying interface renders Markdown and syntax-highlighted code, shows diffs and tool usage, and includes analytics for message distribution, file operations, and conversation statistics. Conversations can be exported as JSON, HTML, or plain text. The UI includes themes, font choices, favorites, and responsive layouts.

## Limitations and notes

The MCP workflow depends on Claude Code history being present in the configured projects directory. The default location is `~/.claude/projects`; use `PROJECT_ROOT` or `--project-root` when histories are stored elsewhere. The web application and MCP server are related parts of the package, but the browser interface is not required for MCP use.

_Full upstream README: https://allmcps.com/mcp/kunwar-shah-claudex/readme_