Lucid Apple MCP — Use Cases

Apple Intelligence is already built into every modern Mac. It's a set of AI tools Apple embedded directly into macOS — and they run entirely on your chip, without sending anything to the internet. We built Lucid Apple as an MCP server that connects those built-in tools directly to Claude Desktop and any local AI app that supports MCP. That means your Mac handles the heavy lifting — reading text from images, reconstructing tables from scanned documents, summarizing, extracting structured data, classifying content — for free, in private, without burning a single API token.

Tools included

ocr — Apple Vision pdf_text — PDF text layer detect — NSDataDetector extract — Apple Intelligence classify — Apple Intelligence summarize — Apple Intelligence generate — Apple Intelligence recognize_document — structured OCR

Local LLM users

Give Ollama / LM Studio vision

OCR

Read tables & forms a text model can't see

recognize_document

Shrink docs past context limit

summarize

Route to right specialized model

classify

Claude users — token savings

OCR screenshot → send text not image

OCR

Tables/forms → structured text, not pixels

recognize_document

Summarize before long analysis

summarize

Pre-classify to skip reasoning tokens

classify

Business & productivity

Business card to contact

OCRextractdetect

Receipt / invoice to expense entry

recognize_documentextract

Meeting notes to action items

extractdetect

Email triage by priority

classifysummarize

OSINT & journalism

OCR documents leaked as images

OCR

Rebuild tables from leaked filings

recognize_document

Extract names, orgs, locations

extract

Classify reports by topic or threat level

classify

Legal & medical

Scanned contracts & records → structured

recognize_documentOCR

Extract parties, dates, obligations

extractdetect

Classify patient messages by urgency

classify

Summarize depositions and clinical notes

summarize

poliLucid pipeline

OCR rep info from web screenshots

OCR

Extract name, title, party, district

extract

Detect office phone and address

detect

Summarize bills and legislation

summarize

Power combos — tool chains

Business card OCR→ extract→ detect→ CRM-ready contact

Scanned contract recognize_document→ extract→ clauses + tables, structured

Document triage summarize→ classify→ route to right workflow

Email → calendar detect→ extract→ structured event JSON

poliLucid scrape OCR→ extract→ detect→ rep record to database

Token savings at scale — illustrative estimates

Modeled scenarios, not customer data — what an image-heavy AI workload could save by OCR-ing locally before sending text (not pixels) to a cloud model.

Company A

Productivity · image-heavy

A writing/productivity assistant where users upload screenshots and PDFs for summarization and analysis — one of the highest-volume image-input workloads a cloud model sees.

$6,750

Without OCR / mo

$1,125

With OCR / mo

$5,625

Saved / mo

Model: 50,000 image-based requests/day · 1,500 avg image tokens → 250 avg OCR tokens · Claude Sonnet at $3/M input

Company B

Legal Document AI

Scanned contracts, court filings, and discovery processed at scale — a heavily image-driven pipeline where each unreadable PDF page arrives as a raw image-token block.

$3,240

Without OCR / mo

$630

With OCR / mo

$2,610

Saved / mo

Model: 20,000 scanned legal pages/day · 1,800 avg image tokens → 350 avg OCR tokens · Claude Sonnet at $3/M input

Lucid Systems · Flagship product

Voical

Lucid Apple is free because we want your Mac doing more for you, privately — but it's a preview of how we build. Voical is the whole idea, assembled: a fully-offline AI suite for Mac — local LLM chat, zero-shot voice cloning in your own voice, document & presentation generation, and a memory that genuinely gets smarter about you — all running on your own silicon. No cloud. No API keys. No subscription. A one-time purchase you own. The free tools show what your Mac can do; Voical is what it becomes when every piece works together. Just as clever as the giveaways — and then some.

100% offline Local LLM chat Voice cloning Document & slide studio Memory that's smarter about you One-time purchase

Coming soon — join the waitlist

Get involved

GitHub — Lucid Apple MCP

Free and open source. Star it, fork it, contribute — or use it as the foundation for your own MCP tools. Every issue filed and PR merged makes it better for the entire local AI community.

View on GitHub

Patreon — Support the Mission

Lucid Apple MCP is free — no license, no subscription, no catch. If it saves you real time and real money, consider supporting Lucid Systems on Patreon. Voluntary. Cancel anytime. Every dollar funds the next tool.

Support on Patreon

Community Project · Open Source · Free Forever

poliLucid — Your Complete Chain of Representation

Every American has roughly 120–160 elected officials making decisions that directly affect their quality of life — from the city council seat two blocks away to the U.S. Senate. Most people can't name five of them. poliLucid is building the most comprehensive open-source database of US elected officials ever assembled. Every level of government. Every state. Every territory. Every seat. We need your help to get there.

We're sourcing detailed records on every elected representative in the country: full contact information, office locations, district boundaries, committee assignments, voting records, campaign finance data, and the political logistics that connect citizens to their government. If you're a developer, researcher, data contributor, or just someone who believes in civic transparency — this is your project too.

Federal · State · Municipal All 50 States + Territories Contact & Office Info Voting Records Committee Assignments District Maps Campaign Finance Community Sourced

Visit poliLucid.com

Getting started — quick answers

What is an MCP server?

MCP stands for Model Context Protocol — an open standard created by Anthropic that lets AI assistants connect to external tools and capabilities. Think of it like a plugin system for AI: once you install an MCP server, your AI can use whatever tools that server provides. Lucid Apple is an MCP server that hands Claude Desktop (and compatible local AI apps) direct access to the AI features already built into your Mac — no extra subscriptions, no cloud accounts, no data leaving your machine.

How do I add Lucid Apple to Claude Desktop?

Claude Desktop reads a config file at startup to know which MCP servers to load. You only need to do this once. Open (or create) this file:

~/Library/Application Support/Claude/claude_desktop_config.json Add the following entry inside the "mcpServers" block — replacing the path with wherever you cloned or downloaded the server:

"lucid-apple": {
  "command": "/usr/local/bin/node",
  "args": ["/path/to/lucid-apple-mcp/server.mjs"]
}

Save the file and restart Claude Desktop. The tools — ocr, pdf_text, detect, extract, classify, summarize, generate, recognize_document — will appear automatically in your next conversation. To find your node path, open Terminal and run: which node

Can I use it with local LLMs like Ollama or LM Studio?

Yes — if your local LLM client supports MCP. Ollama by itself doesn't include an MCP client, but several frontends do: Open WebUI, AnythingLLM, and Continue.dev all support MCP server connections. Add the same config entry in your client's MCP settings section and Lucid Apple's tools become available to any model you run locally. This is the big unlock — it gives vision and document-understanding capabilities to local models that don't natively support images.

What's the difference between OCR and recognize_document?

ocr pulls raw text off an image, line by line — perfect for receipts, signs, screenshots, and prose. recognize_document uses Apple Vision's document recognizer to preserve structure: it returns the full transcript plus tables reconstructed as rows and cells. Use it whenever layout matters — tables, forms, invoices, multi-column pages — where plain OCR would flatten everything into a jumble of words. Both also take PDFs (rasterized automatically). Neither requires Apple Intelligence and both run fully offline — though recognize_document needs macOS 26 (Vision's newer document recognizer), while ocr runs on any Apple Silicon Mac.

Can it read PDFs?

Yes — hand any of the file tools a .pdf path. pdf_text pulls a born-digital PDF's embedded text layer instantly (no OCR, runs on any Apple Silicon Mac). ocr rasterizes and reads scanned PDFs — every page, or one via page. recognize_document rasterizes a page and returns its tables and structure. No extra setup; rasterization happens in memory, nothing is written to disk.

Do I need an Apple Intelligence-enabled Mac?

For ocr, pdf_text, and detect, no — those use Apple Vision, PDFKit, and NSDataDetector and do not require Apple Intelligence. They run on any Apple Silicon Mac (M1 or later). For generate, summarize, extract, and classify, yes — those use Apple's on-device Foundation Models, which require an Apple Silicon Mac (M1 or later) running macOS 26 or later with Apple Intelligence enabled in System Settings. recognize_document is in between: it requires no Apple Intelligence, but it does need macOS 26 (it uses Vision's newer document recognizer). ocr, pdf_text, and detect run on any Apple Silicon Mac regardless of macOS version.