Connects Claude to Apple's on-device 3B parameter model that ships with Apple Intelligence on M-series Macs running macOS 26+. Exposes five tools: fm_generate for text completion, fm_summarize for bullet points, fm_classify for categorization, fm_extract for structured data, and fm_status to check availability. Everything runs locally on the Neural Engine with sub-second to few-second latency and zero API costs. Reach for this when you want fast, private inference for simple tasks like quick classification or extraction where Claude would be overkill, or when you're doing privacy-sensitive work that needs to stay on-device. Requires Apple Silicon and Apple Intelligence enabled in System Settings.
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.
pip install apple-fm-mcp
Add to your Claude Code config (~/.claude/mcp.json):
{
"mcpServers": {
"apple-fm": {
"command": "apple-fm-mcp"
}
}
}
Or Claude Desktop (~/Library/Application Support/Claude/claude_desktop_config.json):
{
"mcpServers": {
"apple-fm": {
"command": "apple-fm-mcp"
}
}
}
| Tool | Description | Latency |
|---|---|---|
fm_generate | Free-form text generation | ~1s |
fm_summarize | Summarize text into bullet points | ~4s |
fm_classify | Classify text into categories | ~2s |
fm_extract | Extract structured data from text | ~3s |
fm_status | Check model availability | instant |
| Task | Use Apple FM | Use Claude |
|---|---|---|
| Quick classification | Yes | Overkill |
| Text summarization | Yes (short text) | Yes (long documents) |
| Code generation | No | Yes |
| Complex reasoning | No | Yes |
| Data extraction | Yes (simple) | Yes (complex) |
| Privacy-critical | Yes | Depends |
AI education from a 501(c)(3) nonprofit. 50+ free courses on Claude, MCP, agents, and AI architecture.
MIT