# aitytech/agentkits-memory [Health: Active]

**Category:** 🧠 Knowledge & Memory  
**Repository:** https://github.com/aitytech/agentkits-memory  
**GitHub Stars:** 64  
**npm Downloads (last month):** 249  
**Views:** 4  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/aitytech-agentkits-memory

## Description
Persistent memory for AI coding assistants with hybrid search (FTS5 + vector embeddings), session tracking, automatic context hooks, and web viewer. SQLite-based with no daemon process — works with Claude Code, Cursor, Windsurf, and any MCP client.

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

```json
"mcpServers": {
  "agentkits-memory": {
    "command": "npx",
    "args": ["-y","@aitytech/agentkits-memory"]
  }
}
```

## Documentation

## What aitytech/agentkits-memory MCP server does

The aitytech/agentkits-memory MCP server adds persistent memory to AI coding workflows. It stores memories, sessions, observations, and embeddings in a local SQLite database, typically at `.claude/memory/memory.db`. Stored information can include architectural decisions, recurring code patterns, error resolutions, file activity, and summaries from earlier sessions.

The system is designed for coding assistants that otherwise lose context when a session ends. On a later session, relevant history can be injected automatically, and the assistant can query the database through MCP instead of relying on manually maintained notes. A browser-based viewer is also available for browsing, searching, creating, editing, and deleting memories.

## How it works

Hooks capture session events, tool activity, decisions, and summaries while work is in progress. Background workers can enrich observations with generated summaries, create local vector embeddings, compress older data, and archive or clean up historical sessions. These tasks run asynchronously rather than blocking the main coding interaction.

Search uses multiple layers. Full-text indexing supports direct term matching through SQLite FTS5, while sqlite-vec and a local multilingual-e5-small ONNX model support semantic similarity across more than 100 languages. A progressive retrieval flow can begin with a compact search result, continue to timeline context, and fetch full details only when needed.

The aitytech/agentkits-memory MCP server does not require a database daemon. Data remains on the local machine, and the README states that no cloud account, external API key, or paid service is required. Hooks are short-lived processes, while worker processes have timeout, locking, signal-handling, and cleanup behavior described by the project.

## Setup and configuration

The package provides an automated setup command:

```bash
npx @aitytech/agentkits-memory
```

It detects supported environments and configures the MCP server, platform-specific hooks or rules, and the embedding model. Platform selection can be explicit:

```bash
npx @aitytech/agentkits-memory --platform=cursor
npx @aitytech/agentkits-memory --platform=windsurf,cline
npx @aitytech/agentkits-memory --platform=all
```

The documented integrations include Claude Code, Cursor, Windsurf, Cline, and OpenCode. Claude Code also has a plugin-marketplace installation path that adds the hooks, MCP server, and memory workflow skill. The web viewer starts with:

```bash
npx @aitytech/agentkits-memory web
```

After starting it, open `http://localhost:1905`. The project does not list environment variables or credential configuration as required setup.

## Tools and capabilities

The MCP interface exposes nine listed operations:

- Save a memory
- Search memories
- Retrieve timeline context
- View memory details
- Recall stored context
- List memories
- Update a memory
- Delete a memory
- Check system status

Additional capabilities include automatic context capture, multilingual local vector search, memory namespaces and tags, session timelines, JSON export and import, and lifecycle management for older data. The web viewer provides management controls separate from the MCP tools.

## Limitations and notes

This project is aimed at local coding-assistant memory rather than a hosted, shared memory service. Its data is stored in a local SQLite file, so synchronization between machines or users is not described. Semantic search depends on the locally downloaded embedding model. The README states that background processing performs AI enrichment, but it does not identify an external model provider or require an API key.

The documented client list is broader than the four clients commonly used for MCP directory compatibility: Claude Code, Cursor, Windsurf, Cline, and OpenCode are named explicitly. Other MCP clients may work, but the supplied material does not provide client-specific configuration details for them.

_Full upstream README: https://allmcps.com/mcp/aitytech-agentkits-memory/readme_