Némos
NémosBack to Home

Developer · Model Context Protocol

Némos MCP

Give any AI agent live access to your Némos second brain. The Némos MCP server lets Claude, ChatGPT, Gemini, Cursor — any host that speaks the Model Context Protocol — search, enrich, and reason over your screenshots, notes, links, reels, and PDFs. One Apple sign-in. No per-provider API keys.

At a glance

  • Connect any MCP host — Claude, ChatGPT, Gemini, Cursor — to your Némos library. One Apple/CloudKit sign-in is the only credential.
  • Runs locally over stdio with npx -y nemos-mcp, or hosted over HTTPS for cloud agents.
  • 36 tools across four layers — retrieve, enrich, connect, deliver — with build_context_pack as the keystone.
  • Your data never touches Némos servers; the agent reads your private CloudKit directly.

How it works

The MCP server owns one thing: a secure connection to your private CloudKit data. The AI host does the reasoning. The agent follows a simple loop — retrieve from your library, enrich only the gaps, connect related items, and deliver a context pack you can actually use.

Because it speaks a single open protocol, the same server works in every AI tool — you don't build a separate integration per provider. Your data connection is the only external link; everything else runs locally or on a server you control.

Install the MCP server

You need Node.js 20+ and an MCP-capable host. No clone, no build — the server runs straight from npm. Pick your client:

Run this in your project root (or drop the JSON into a .mcp.json file):

claude mcp add nemos -- npx -y nemos-mcp

Sign in once

  1. Restart your AI host so it loads the nemos server, then ask it to connect to Nemos (tool nemos_connect).
  2. A browser opens to Apple's iCloud sign-in. Sign in with the same Apple ID you use in the Némos app and complete 2FA.
  3. You're connected. The token is stored in your OS keychain and refreshes itself — you won't sign in again.
A note on the sign-in screen. Apple's consent page may name the app by Némos's internal iCloud container (iCloud.dev.codewithbeto.screennameapp) rather than "Némos". That is the correct, safe Némos container — the same one the app itself uses. The friendly "Némos" name appears once the app is published on the App Store.

Hosted (remote)

For cloud hosts that can't spawn a local process (ChatGPT or Gemini web), run one shared HTTP server. Each user sends their own CloudKit token in a header — sessions are isolated and the token never touches disk.

{
  "mcpServers": {
    "nemos": {
      "type": "http",
      "url": "https://mcp.nemosapp.com/mcp",
      "headers": { "X-Nemos-CK-Token": "YOUR_CK_WEB_AUTH_TOKEN" }
    }
  }
}

Tools

36 tools across four layers. Badges mark tools that need an optional AI key or yt-dlp on the server.

Authentication

One Apple/CloudKit sign-in — the only external connection the agent needs.

nemos_connectlocal only

Connect in one step: opens Apple sign-in, captures the token, verifies.

nemos_sign_in_url

Get the Apple CloudKit sign-in URL to open in a browser.

nemos_set_token

Store the ckWebAuthToken (paste the callback URL or raw token).

nemos_auth_status

Check whether the stored token is valid.

ensure_tools

Report (and optionally install) local enrichment tools like yt-dlp.

Layer 1 · Retrieve

Read straight from your library. Every saved item already carries extracted text.

search_nemos

Search saved items by text across the whole library.

get_item

Get one item in full by id.

list_folders

List all folders.

get_folder_items

Items inside a folder.

get_recent_items

Most recently updated items.

get_favorites

Favorited items.

get_related_items

Items related to a seed item.

get_reminders

All reminders.

get_conversations

All chat conversations.

get_smart_collections

All smart collections.

get_asset

Download one asset (file or thumbnail) of an item.

download_item_files

Download every file attached to an item.

export_all

Bulk-export every synced item's files plus a manifest.

Layer 2 · Enrich

Fill the gaps — only fetch what the stored text is missing. Cached and idempotent.

fetch_url_content

Fetch raw HTML/text of a URL.

extract_article_text

Clean article text from a URL (readability extraction).

summarize_articleneeds AI key

Summarize an article at a URL.

get_youtube_metadataneeds yt-dlp

YouTube video metadata.

get_youtube_transcriptneeds yt-dlp

YouTube transcript from auto-captions.

get_reel_metadataneeds yt-dlp

Instagram / TikTok reel metadata.

get_reel_transcript_or_captionneeds yt-dlp

Reel transcript if available, else caption.

extract_pdf_text

Extract text from a saved PDF item.

summarize_pdfneeds AI key

Summarize a saved PDF item.

extract_ocr_from_image

OCR a saved image or screenshot item.

analyze_image_or_screenshotneeds AI key

Vision analysis of an image item.

Layer 3 · Connect & Deliver

Surface structure across your library and assemble it into something usable.

link_items

Link two saved items with a typed reason.

find_related_by_topic

Find saved items related to a topic query.

find_duplicate_items

Find duplicate / near-duplicate saved items.

generate_knowledge_map

Return nodes + edges for a topic (graph structure).

export_to_obsidian

Write a context-pack note into an Obsidian vault.

create_research_briefneeds AI key

Deep research brief for a topic.

build_context_packneeds AI key

Keystone: search → enrich gaps → synthesize into a knowledge pack.

Example prompts

Once Némos is connected, paste any of these into your AI host.

“Find everything I saved about AI agents and build me a research brief.”
“What did I save this week? Group it by theme and summarize.”
“I saved a bunch of articles on one topic — find the duplicates and tell me which three actually matter.”
“Pull the transcript from that YouTube video I saved and give me the five key takeaways.”
“Turn my saved screenshots about this project into an Obsidian note.”
“Find my reminders and notes about this event and draft a plan from them.”

Best practices

  • Mention Némos explicitly so the agent reaches for your library before its own guesses.
  • Start from retrieval — ask it to search your saved items first, then reason.
  • For anything that spans multiple sources, use build_context_pack.
  • Let it enrich only the gaps — it reuses stored text and fetches just what's missing.
  • Be specific about output: a summary, an Obsidian note, a research brief, or a PRD.
  • Connect, don't just collect — ask it to link related items and flag duplicates.

Privacy

The MCP server connects to your own private CloudKit database with a rotating web-auth token — the same mechanism Némos itself uses. Nothing is sent to Némos servers. Run it locally and your data never leaves your machine; run it hosted and you control the host. Questions? [email protected].