Spark
Sign In
MCP Connector

Generate Text with On-Device Foundation Models

Runs Apple's on-device Foundation Models framework as an MCP server, for private, local text generation on Apple Silicon Macs.

Connect
Works with apple
Eric Marchand
Own this? Claim it

Maintainer of this project? Claim this page to edit the listing.

77
Spark score
out of 100
Updated Jun 2025
Fresher alternatives ↓
Version 1.0.0
Models
universal
Add to Favorites

Why it matters

Leverage Apple's on-device Foundation Models for secure and private text generation. This MCP server provides local access to language models, ideal for macOS applications requiring AI capabilities without external dependencies.

Outcomes

What it gets done

01

Integrate with Apple's Foundation Models framework for text generation.

02

Provide secure, private, and on-device language model access.

03

Enable text generation for macOS applications on Apple Silicon.

04

Support system instructions for customized AI responses.

Install

Add it to your toolbox

Run in your project directory:

curl -fsSL https://spark.entire.vc/get/vb-foundationmodels | bash

Capabilities

Tools your agent gets

generate_text

Generate text using Apple's Foundation Models framework with secure on-device processing.

Overview

FoundationModels MCP Server

This MCP server exposes Apple's on-device Foundation Models framework for private, local text generation, configurable via default system instructions. Use it on Apple Silicon Macs running macOS 26.0+ for private, local text generation. Conversation history/session management is not yet implemented.

What it does

mcp-foundation-models is an MCP server that exposes text generation through Apple's Foundation Models framework, letting MCP clients access Apple's on-device language models for secure, private text generation that runs entirely locally rather than through a cloud API.

When to use - and when NOT to

Use this when you want an MCP client to generate text using Apple's on-device models directly on a Mac, keeping generation private and local rather than routing through a third-party API. It requires macOS 26.0 (Tahoe) or later, Xcode 26.0+, Swift 6.2+, and an Apple Silicon Mac for optimal Foundation Models performance - it will not run on older macOS versions or Intel Macs without those requirements met. Session/conversation history management is listed as a to-do item, not yet implemented, so don't expect multi-turn context to persist automatically across calls.

Capabilities

Text generation via Apple's Foundation Models framework, configurable with SYSTEM_INSTRUCTIONS (default system prompt) and DEBUG (enable debug logging) environment variables.

How to install

git clone <repository-url>
cd mcp-foundation-models
swift build -c release

The executable is produced at .build/release/mcp-foundation-models. Claude Desktop configuration:

{
  "mcpServers": {
    "foundation-models": {
      "command": "/path/to/mcp-foundation-models/.build/release/mcp-foundation-models",
      "args": []
    }
  }
}

Who it's for

macOS developers on Apple Silicon who want private, on-device text generation through Apple's Foundation Models framework instead of a cloud-based LLM API. Licensed MIT.

Source README

mcp-foundation-models

A Model Context Protocol (MCP) server that provides text generation capabilities using Apple's Foundation Models framework. This server enables MCP clients to access Apple's on-device language models for secure, private text generation.

Features

  • Apple Foundation Models Integration: Leverages Apple's on-device language models for text generation

Requirements

  • macOS 26.0 or later (macOS Tahoe)
  • Xcode 26.0 or later
  • Swift 6.2 or later
  • Apple Silicon Mac (for optimal Foundation Models)

Installation

Building from Source

  1. Clone the repository:

    git clone <repository-url>
cd mcp-foundation-models

  2. Build the project:

    swift build -c release

  3. The executable will be available at:

    .build/release/mcp-foundation-models

Use it in Claude Desktop for instance

Edit Claude configuration file '$HOME/Library/Application Support/Claude/claude_desktop_config.json'

Add this server full path as "mcpServers" sub object 

{
  "mcpServers": {
    "foundation-models": {
      "command": "/path/to/mcp-foundation-models/.build/release/mcp-foundation-models",
      "args": [
      ]
    }
Environment Variables

The server supports configuration through environment variables:

  • SYSTEM_INSTRUCTIONS: Set default system instructions for the AI assistant
  • DEBUG: Enable debug logging (any non-empty value)

Dependencies

TODO

  • manage some session id to keep some conversation history

FAQ

Common questions

Discussion

Questions & comments · 0

Sign In Sign in to leave a comment.