MCP.so
登录

M1K3 MCP Server已验证已精选

@Round-Tower

基本信息

分类

记忆与知识

许可证

NOASSERTION

作者

Round-Tower

提交者

Kevin Murphy

关于 M1K3 MCP Server

Local AI agent for your Mac with a built-in MCP server: 18 tools for your notes and PDFs, a temporal memory graph, an on-device LLM and a neural voice. No account, no telemetry. No data centers required.

M1K3 MCP Server 配置

使用下面的配置,将此服务器添加到你的 MCP 客户端。

{
  "mcpServers": {
    "m1k3": {
      "type": "http",
      "url": "http://127.0.0.1:4242/mcp",
      "headers": {
        "Authorization": "Bearer <token from M1K3 > Settings > MCP Server>"
      }
    }
  }
}

M1K3 MCP Server 工具

工具列表暂未提供

当前收录暂时没有可展示的工具列表,不代表服务没有工具。请查阅项目文档了解支持的能力。

点击上方按钮,直接连接服务器端点实时获取工具列表。

概览

What is M1K3?

M1K3 ("Mike") is a local AI agent for Apple Silicon Macs from Round Tower. It chats with an on-device language model, talks with on-device speech recognition and a neural voice, answers from the notes and PDFs you drop in, and remembers facts in a temporal memory graph you can see, search and delete. No account, no telemetry. No data centers required.

Its built-in MCP server gives Claude Code, Codex, Cursor, VS Code and Zed the same brain, memory and voice: 18 tools over loopback HTTP, guarded by an access token.

How to use M1K3?

  1. Install M1K3 from the Mac App Store or Homebrew (brew trust round-tower/tap && brew install --cask round-tower/tap/m1k3).
  2. In the app, turn on Settings > MCP Server and copy the access token.
  3. In Terminal: m1k3 login && m1k3 connect claude (also codex, cursor, vscode, zed). It asks for the token with echo off and writes your client's config.

Or add it by hand. JSON clients take the server configuration on this page; for Claude Code:

claude mcp add --transport http -s user m1k3 http://127.0.0.1:4242/mcp \
  --header "Authorization: Bearer m1k3_..."

Key features of M1K3

  • Your own knowledge: search_knowledge, list_documents and get_document over the notes, PDFs and calls M1K3 has indexed (hybrid retrieval).
  • A memory graph: remember, recall_memory, related_memory and forget_memory. Dated facts that supersede each other instead of piling up.
  • A second brain on-device: ask_m1k3 asks M1K3's local LLM; long answers come back as jobs (get_answer, list_jobs).
  • A voice: speak, stop_speaking and listen, so an agent can tell you when it is blocked or done, out loud.
  • Todos and links: propose_todo, list_todos, open_link.

Use cases of M1K3

  • Give a coding agent a memory of your project decisions that outlives the session.
  • Let an agent answer from your own PDFs and notes without uploading them anywhere.
  • Hear a long-running agent's milestones and blockers aloud instead of watching a terminal.

常见问题

1

Does anything leave my Mac?

Conversations, notes and documents stay on the device by default. M1K3 uses the network to download models when you ask, and for web search, which is on by default and one switch in Settings to turn off.

2

What does it need?

macOS 26 or later on Apple Silicon. The MCP server runs while the app is open.

3

Is it open source?

It is source-available under FSL-1.1-ALv2: the code is on GitHub, and each release becomes Apache-2.0 after two years.

评论

记忆与知识 分类下的更多 MCP 服务器