Notion

Connect Crow to Notion to search, read, and create pages and databases through your AI assistant.

What You Get

• Search across your Notion workspace
• Read and create pages
• Query and update databases
• Add and read comments on pages

Setup

Step 1: Create an internal integration

1. Go to notion.so/my-integrations
2. Click New integration
3. Name it (e.g., "Crow") and select your workspace
4. Under Type, keep Internal selected
5. Click Save
6. Copy the Internal Integration Secret (starts with ntn_)

Step 2: Share pages with the integration

Notion integrations can only access pages that have been explicitly shared with them:

1. Open a Notion page or database you want Crow to access
2. Click the ... menu in the top-right corner
3. Click Connections → Connect to → find and select your integration
4. Repeat for each page or database

Child pages inherit the connection, so sharing a top-level page grants access to all its subpages.

Step 3: Add to Crow

Paste your integration secret in Crow's Nest → Settings → Integrations, or on the Setup page at /setup.

The environment variable is NOTION_TOKEN.

Required Permissions

Permissions are configured on the integration settings page at notion.so/my-integrations:

Permission	Why
Read content	Search and read pages and databases
Update content	Edit existing pages and database entries
Insert content	Create new pages and database entries
Read comments	View comments on pages
Insert comments	Add comments to pages

Troubleshooting

"object_not_found" error

The page or database hasn't been shared with your integration. Open the page in Notion, click ... → Connections, and add your integration.

Can't find pages in search

Notion search only returns pages that have been explicitly connected to the integration. Share the parent page to grant access to all subpages.

Token starts with "secret_" instead of "ntn_"

Older Notion integrations used the secret_ prefix. Both formats work — just paste the full token as-is.

Sync Notion into Crow memory (local semantic search)

Notion's API (and its MCP server) only does keyword search. Crow's semantic search runs over the memories table, not over Notion live. scripts/sync-notion.js bridges the two: it pulls every page shared with your integration, converts it to Markdown, and stores each as a memory (category=learning, tags=notion,sync). The normal embedding pipeline then makes that content semantically searchable through crow_search_memories / crow_deep_recall from any connected client.

It's idempotent — pages are deduped on source = notion:<pageId> and only re-embedded when their Notion last_edited_time changes — and inert when NOTION_TOKEN is unset, so instances that don't use Notion are unaffected.

Run it

# Token must be in the environment (e.g. via --env-file=.env, Node >= 20):
npm run sync-notion -- --once --dry-run --limit 5   # preview decisions, no writes
npm run sync-notion -- --once                        # full sync
npm run sync-notion -- --once --force                # re-embed everything

Flags: --dry-run (show insert/update/skip, no writes), --limit N (cap pages), --force (ignore last_edited_time). If the embedding provider is offline the pages are still stored (keyword-searchable); run node scripts/backfill-embeddings.js --only memories later to fill embeddings.

One memory per page (v1). Very long, multi-topic pages are averaged into a single embedding and content past ~8000 chars is excluded from the embedding (still stored and keyword-searchable). The source key already supports a future per-section (notion:<pageId>#<n>) chunking upgrade with no schema change.

Schedule it

Run the sync on a timer. A macOS launchd template is provided at examples/notion-sync/com.crow.notion-sync.plist.example (fill in the placeholder paths, copy to ~/Library/LaunchAgents/, launchctl load).

Why not crow_create_schedule?

The built-in scheduler only advances DB rows and emits notifications — it does not spawn external scripts. Use an OS timer (launchd / systemd / cron) instead.

Linux (systemd user timer) — ~/.config/systemd/user/crow-notion-sync.{service,timer}:

# crow-notion-sync.service
[Service]
Type=oneshot
WorkingDirectory=%h/Developer/crow
ExecStart=/usr/bin/node --env-file=%h/Developer/crow/.env scripts/sync-notion.js --once

# crow-notion-sync.timer
[Timer]
OnBootSec=10min
OnUnitActiveSec=6h
[Install]
WantedBy=timers.target

Enable with systemctl --user enable --now crow-notion-sync.timer.

cron (every 6 hours):

0 */6 * * * cd $HOME/Developer/crow && /usr/bin/node --env-file=.env scripts/sync-notion.js --once >> $HOME/.crow/logs/notion-sync.log 2>&1