Local-first, zero-API memory for AI agents. Persistent, private, and yours — runs entirely on your machine, no cloud calls, no API keys.
概览
No install step — uvx fetches and runs it: "Remember that we chose SQLite over Postgres for this project because it ships in a single file." …then, in a completely new session tomorrow: That's it. Your agent now remembers, and nothing left your laptop. The config above covers most clients, but several agents want a different shape — and get it wrong silently. VS Code's root key is servers; Codex uses TOML; OpenCode and Kilo Code take command as an array; Goose calls them extensions; Zed nests them under context_servers. has the verified config for 20+ agents: Claude Code, Codex, Gemini CLI, Copilot CLI, Goose, OpenCode, Crush, Amp, Amazon Q, Qwen Code, Junie, Antigravity, Warp, Cursor, Windsurf, Zed, VS Code, JetBrains, Trae, Cline, Roo Code, Kilo Code, Continue, and Claude Desktop. Plus memory_stats for where the database lives and how much is in it. The MCP server is a thin shell over a store you can import directly:
README
以下为原文 README 节选，完整内容请查看原文。
Quickstart (30 seconds)
1. Add it to your MCP client. No install step — uvx fetches and runs it:
// Claude Desktop: claude_desktop_config.json
// Cursor: .cursor/mcp.json
// Claude Code: claude mcp add localmem -- uvx localmem-mcp
{
"mcpServers": {
"localmem": {
"command": "uvx",
"args": ["localmem-mcp"]
}
}
}
2. Restart the client and talk to it:
“Remember that we chose SQLite over Postgres for this project because it ships in a single file.”
…then, in a completely new session tomorrow:
“What database did we pick, and why?”
That’s it. Your agent now remembers, and nothing left your laptop.
The three tools
| Tool | What the agent uses it for |
|---|
安装
uvx localmem-mcp配置
// Claude Desktop: claude_desktop_config.json
// Cursor: .cursor/mcp.json
// Claude Code: claude mcp add localmem -- uvx localmem-mcp
{
"mcpServers": {
"localmem": {
"command": "uvx",
"args": ["localmem-mcp"]
}
}
}