M1K3

What M1K3 does

The M1K3 MCP server gives coding agents access to information and capabilities held by the M1K3 Mac application. Its data sources include imported notes and PDFs, a personal knowledge graph, dated memories, and encrypted call transcriptions. The app also provides an on-device AI model that agents can query through ask_m1k3.

M1K3 runs locally on an Apple-Silicon Mac and is designed for workflows where documents, memories, and agent requests remain on the device. Knowledge answers can include citations from locally added material. Web search is enabled by default in the application, but it can be disabled in Settings; the Developer ID DMG does not provide a cloud model path.

How it works

The running M1K3 application acts as the daemon and serves MCP at http://127.0.0.1:4242/mcp. The MCP endpoint is disabled until the user enables it under M1K3 Settings. Requests must include the access token configured in the app, using an Authorization: Bearer header. Requests without the token receive HTTP 401.

A bundled m1k3 command-line client manages authentication and client configuration. m1k3 login stores the token in the system keychain, while m1k3 connect prepares an MCP configuration for a supported agent. The command-line client can also ask the resident model, save memories, search knowledge, speak text, and report status.

Setup and configuration

Install M1K3 from the Mac App Store, Homebrew, a signed DMG, or from source. The documented product requirement is macOS 26 Tahoe or later on Apple Silicon. Open the app, enable the MCP server in Settings, and copy or authenticate with the displayed token.

For a Homebrew installation, the documented sequence is to trust the Round Tower tap, install the cask, run m1k3 login, and then use m1k3 connect with a client such as Claude, Codex, Cursor, Visual Studio Code, or Zed. The App Store helper cannot write another application's configuration file or run its command, so it prints configuration for manual pasting. The Developer ID DMG and Homebrew versions can write the configuration.

Tools and capabilities

The M1K3 MCP server exposes 18 tools generated from the running service. The README identifies capabilities for:

• Asking the local M1K3 model questions with ask_m1k3.
• Searching local knowledge and retrieving sources.
• Accessing documents and stored memories.
• Using voice-related functions.
• Supporting agent workflows grounded in the user's own data.

The README excerpt does not include the complete generated tool table, so individual tool names beyond ask_m1k3 should be checked against the running server.

Limitations and notes

The application must be running for the MCP endpoint to respond. Its MCP service is local rather than a remotely hosted endpoint, and access requires the per-server token. The Mac implementation is the product surface described here; iOS and visionOS builds share parts of the source package but are not documented as MCP hosts in the supplied material.

The server depends on M1K3's local application state, including its imported documents and memories. Larger local models may need to be downloaded when requested. Web search may use the network unless disabled, while the Developer ID DMG has no cloud model path. M1K3 is therefore most suitable for agents running on the same Mac and able to reach the loopback HTTP address.