Instagram Saved → Notion Inspiration Library

Automatically syncs posts from an Instagram saved collection into a Notion Inspiration Library database. Each post gets transcribed (if it's a reel), analyzed by Claude Haiku for hook frameworks and structural patterns, and saved with full metadata.

This runs as a background daemon. You save content on Instagram. It shows up in your Inspiration Library, analyzed and tagged, without you doing anything.

Permissions

What it accesses:

• Your Instagram saved collections (read-only, via instagrapi)
• Your Notion Inspiration Library database (write: creates pages)
• Claude Haiku API (for content analysis, ~$0.01/post)
• yt-dlp + Whisper (local, for reel transcription)

What it does NOT do:

• Post, like, comment, or follow on Instagram
• Delete or modify existing Notion pages
• Access any Instagram data beyond your saved collections
• Store credentials anywhere except the local .env file

Prerequisites

1. Instagram account with a saved collection containing content you want to analyze
2. Notion integration with access to your Inspiration Library database
3. Claude Code installed (npm install -g @anthropic-ai/claude-code)
4. Python 3.10+ with pip

Setup

1. Install Dependencies

cd ~/ai-content-system/skills/ig-saved-sync
pip install -r requirements.txt

2. Configure Environment

cp .env.example .env

Edit .env with your credentials:

IG_USERNAME=your_username
IG_PASSWORD=your_password
COLLECTION_NAME=AI to emulate
NOTION_TOKEN=ntn_your_token
NOTION_DB_ID=your_database_id

Finding your Notion database ID: Open your Inspiration Library in the browser. The URL looks like notion.so/your-workspace/abc123def456?v=.... The abc123def456 part (32 hex characters) is your database ID.

3. First Run (Interactive)

Run status check first to verify everything connects:

python3 ig-saved-sync.py status

Expected output:

Instagram Saved Sync — Status
========================================

IG Username: your_username
Collection: AI to emulate
Notion DB: abc123de...
Processed posts: 0

Testing Instagram login...
  IG login: OK
  Collection 'AI to emulate': FOUND
  Posts in collection: 47
  Unprocessed: 47

Testing Notion API...
  Notion DB: OK (Inspiration Library)

If your account has 2FA enabled, the first login will prompt for a code. Run status interactively to establish the session, then automated runs will reuse it.

4. Test With a Single Post

python3 ig-saved-sync.py backfill --limit 1

Check your Notion database. You should see one new entry with:

• Title (descriptive, from Claude analysis)
• Hook text and framework classification
• Performance tier (Outlier/Strong/Reference/Untested)
• Topic tags
• Structural notes
• Caption and transcript in the page body

Commands

run — Incremental Sync

python3 ig-saved-sync.py run

Processes only new posts added since the last run. Skips anything already in the processed state file. This is what the daily automation runs.

backfill — Import Existing Posts

python3 ig-saved-sync.py backfill
python3 ig-saved-sync.py backfill --limit 10
python3 ig-saved-sync.py backfill --no-analyze

Options:

• --limit N — Only process N posts (useful for testing or rate limit concerns)
• --no-analyze — Skip Claude analysis, use basic metadata only (faster, cheaper)

status — Check Connections

python3 ig-saved-sync.py status

Verifies Instagram login, collection access, and Notion API connectivity. Shows how many posts are processed vs unprocessed.

Architecture

Instagram Saved Collection
      │
      ▼
┌─────────────┐     ┌──────────────┐     ┌────────────┐
│ Fetch Posts  │────▶│  Transcribe  │────▶│  Analyze   │
│ instagrapi   │     │  yt-dlp +    │     │  Claude    │
│              │     │  whisper     │     │  Haiku     │
└─────────────┘     └──────────────┘     └─────┬──────┘
                                                │
                                                ▼
                                         ┌────────────┐
                                         │   Notion   │
                                         │ Inspiration│
                                         │  Library   │
                                         └────────────┘

Flow per post:

1. Fetch — instagrapi pulls the post metadata (caption, engagement, creator, URL)
2. Transcribe — For reels/videos under 90s, yt-dlp downloads audio, Whisper transcribes locally
3. Analyze — Claude Haiku receives caption + transcript + metrics, returns structured JSON
4. Save — Creates a Notion page with all properties + caption/transcript in the body

If transcription fails, analysis still runs using caption + metadata only. If analysis fails, a basic fallback entry is created from the raw metadata.

Analysis Output

Each post gets classified into:

Field	Example
title	"Contrarian ChatGPT Usage Pattern"
type	Hook Example, Creator Pattern, Script Format, Carousel Example
hook_text	"Stop using ChatGPT like this..."
hook_framework	Curiosity Gap, Proof-First, Pain Point, Contrarian, etc.
performance_tier	Outlier (100K+ views), Strong (10K+), Reference, Untested
tags	["AI Tools", "Tutorial", "Productivity"]
notes	"Opens with contrarian statement. Rapid demo of wrong vs right."

Automation

macOS (launchd)

Copy the template and customize:

cp launchd-template.plist ~/Library/LaunchAgents/com.user.ig-saved-sync.plist

Edit the path in the plist to match your install location, then load it:

launchctl load ~/Library/LaunchAgents/com.user.ig-saved-sync.plist

This runs the sync daily at 11pm. Check logs at /tmp/ig-saved-sync-stderr.log.

To stop:

launchctl unload ~/Library/LaunchAgents/com.user.ig-saved-sync.plist

Linux (cron)

crontab -e

Add:

0 23 * * * cd ~/ai-content-system/skills/ig-saved-sync && python3 ig-saved-sync.py run

State Management

The script tracks which posts have been processed in state/processed.json. Each entry maps an Instagram media PK to its Notion page ID and sync timestamp.

This means:

• Running run multiple times is safe. It skips already-processed posts.
• If you delete a Notion page, the script won't recreate it (it still shows as processed).
• To reprocess a post, remove its entry from state/processed.json.
• To start fresh, delete state/processed.json entirely.

Session data (Instagram login cookies) is stored in state/session.json. This avoids re-authentication on each run. Sessions typically last 2-4 weeks before needing a refresh.

Security

• All credentials are stored in .env (never committed to git)
• Instagram session cookies are in state/session.json (local only)
• Claude analysis uses --max-budget-usd 0.05 per call (cost cap)
• The script uses --permission-mode bypassPermissions for Claude since it only needs text analysis (no file access, no tools)
• No data is sent anywhere except Notion and the Claude API
• .env, state/, and logs/ should all be in your .gitignore

Troubleshooting

"2FA required but running non-interactively" Run python3 ig-saved-sync.py status in a terminal to refresh the session with 2FA.

"Collection not found" The collection name in .env must match exactly (case-sensitive). Run status to see available collection names.

"Notion API error 400" Usually means a property name in the script doesn't match your database. Check that your Inspiration Library has these properties: Title, Type, Source, Platform, Date Added, Source URL, Creator Handle, Performance Tier, Engagement Metric, Topic Tags, Hook Framework, Hook Text, Notes.

Transcription fails silently Make sure yt-dlp and whisper are installed. Test with: python3 -m yt_dlp --version and python3 -m whisper --help.

Rate limiting from Instagram Increase RATE_LIMIT_SECONDS in .env. Default is 5 seconds between posts. For backfills, it uses 10 seconds. If you're getting blocked, try 15-20.