apple-fm-mcp

io.github.sophiacavePublished by GitHub account sophiacavev1.0.1Updated Oct 27, 2026

Apple Foundation Models on-device. Generate, summarize, classify via Apple Intelligence.

VerifiedSTDIODesktop onlyKnowledge & MemoryAI & ML

Overview

AI-generated overview

Exposes Apple's on-device Foundation Model to Claude via MCP for free, private text generation, summarization, classification and extraction on a Mac.

What you can do
  • Generate free-form text with a local on-device model
  • Summarize text into bullet points
  • Classify text into categories
  • Extract structured data from text
  • Check whether the on-device model is available
When to use it
Useful on Apple Silicon Macs when you want quick, low-cost, privacy-preserving classification, short-text summarization or simple data extraction without sending content to a cloud API. It complements Claude rather than replacing it for code generation or complex reasoning.
Requirements
Runs as a local Python package (pip install apple-fm-mcp) started over stdio. Needs macOS 26 or later, Apple Silicon (M1 or newer), Apple Intelligence enabled in System Settings, and Python 3.10 or newer. No API keys or environment variables are declared.
Before you install
Desktop-only and limited to recent Apple hardware and OS versions; it will not work on Intel Macs or older macOS. Inference is local, so text is not sent to a third party, but the model is small and suited only to simple tasks.

Installation

In SourceWeft

  1. Open apple-fm-mcp in the dashboard and add it to a workspace.
  2. Enable the server for the chats that should use its tools.

Desktop only via STDIO. STDIO servers start a local process, so they need the SourceWeft desktop host.

Other MCP clients

Follow the launch instructions in the repository.

README

Apple Foundation Models MCP Server

Expose Apple's on-device 3B parameter LLM to Claude Code and Claude Desktop via MCP. Zero API cost, fully private, runs on Neural Engine.

Requirements

  • macOS 26+ (Tahoe or later)
  • Apple Silicon (M1/M2/M3/M4)
  • Apple Intelligence enabled in System Settings
  • Python 3.10+

Install

bash
pip install apple-fm-mcp

Configure

Add to your Claude Code config (~/.claude/mcp.json):

json
{  "mcpServers": {    "apple-fm": {      "command": "apple-fm-mcp"    }  }}

Or Claude Desktop (~/Library/Application Support/Claude/claude_desktop_config.json):

json
{  "mcpServers": {    "apple-fm": {      "command": "apple-fm-mcp"    }  }}

Tools

ToolDescriptionLatency
fm_generateFree-form text generation~1s
fm_summarizeSummarize text into bullet points~4s
fm_classifyClassify text into categories~2s
fm_extractExtract structured data from text~3s
fm_statusCheck model availabilityinstant

Why Use This?

  • Free: No API keys, no tokens, no cloud costs
  • Private: All inference on-device, no data leaves your Mac
  • Fast: Neural Engine optimized, ~1s for generation
  • Complementary: Use alongside Claude for tasks where on-device speed matters

When to Use Apple FM vs Claude

TaskUse Apple FMUse Claude
Quick classificationYesOverkill
Text summarizationYes (short text)Yes (long documents)
Code generationNoYes
Complex reasoningNoYes
Data extractionYes (simple)Yes (complex)
Privacy-criticalYesDepends

Built by Like One

AI education from a 501(c)(3) nonprofit. 50+ free courses on Claude, MCP, agents, and AI architecture.

License

MIT

Source: README.md at commit 618bc12

Tools

0
SourceWeft has not started this package in its sandbox yet, so its tools are not known.

Version history

1
  1. v1.0.1LatestSep 16, 2026

Score

40% of the applicable checks met

How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.

  • Reached in the last 30 days25 pts

    Keep the endpoint answering the MCP handshake; SourceWeft probes it every 30 days.

  • Auth declared10 pts

    Declare the credential headers in the registry entry, or keep the endpoint answering the probe so its auth can be classified.

  • Tools listed15 pts

    Answer tools/list without a credential, or ship a package that starts in a sandbox without one.

  • DescriptionRequired15 pts
  • Logo10 pts
  • Documentation15 pts
  • License10 pts

    Declare a license in the repository or the registry entry.