Agent Notion

A TypeScript CLI tool that enables AI agents and humans to interact with Notion workspaces through the unofficial private API. Supports full CRUD operations on pages, databases, blocks, search, and user management.

Note: This skill uses Notion's internal/private API (/api/v3/), which is separate from the official public API. For official API access, use agent-notionbot.

Quick Start

bash

# 1. Extract token_v2 from Notion desktop app
agent-notion auth extract

# 2. Find your workspace ID
agent-notion workspace list --pretty

# 3. Search for a page
agent-notion search "Roadmap" --workspace-id <workspace-id> --pretty

# 4. Get page content
agent-notion page get <page-id> --workspace-id <workspace-id> --pretty

# 5. Query a database
agent-notion database query <collection-id> --workspace-id <workspace-id> --pretty

Important: --workspace-id is required for ALL commands that operate within a specific workspace. Use agent-notion workspace list to find your workspace ID.

Authentication

Token Extraction (Desktop App)

Extract token_v2 from the Notion desktop app automatically. No API keys or OAuth needed.

bash

# Extract token_v2 from Notion desktop app
agent-notion auth extract

# Check auth status (shows extracted token_v2)
agent-notion auth status

# Remove stored token_v2
agent-notion auth logout

On macOS, your system may prompt for Keychain access — this is normal and required to decrypt the cookie.

The extracted token_v2 is stored at ~/.config/agent-notion/credentials.json with 0600 permissions.

Memory

The agent maintains a ~/.config/agent-notion/MEMORY.md file as persistent memory across sessions. This is agent-managed — the CLI does not read or write this file. Use the Read and Write tools to manage your memory file.

Reading Memory

At the start of every task, read ~/.config/agent-notion/MEMORY.md using the Read tool to load any previously discovered workspace IDs, page IDs, database IDs, and user preferences.

•If the file doesn't exist yet, that's fine — proceed without it and create it when you first have useful information to store.
•If the file can't be read (permissions, missing directory), proceed without memory — don't error out.

Writing Memory

After discovering useful information, update ~/.config/agent-notion/MEMORY.md using the Write tool. Write triggers include:

•After discovering workspace IDs (from workspace list)
•After discovering useful page IDs, database IDs, collection IDs (from search, page list, page get, database list, etc.)
•After the user gives you an alias or preference ("call this the Tasks DB", "my main workspace is X")
•After discovering page/database structure (parent-child relationships, what databases live under which pages)

When writing, include the complete file content — the Write tool overwrites the entire file.

What to Store

•Workspace IDs with names
•Page IDs with titles and parent context
•Database/collection IDs with titles and parent context
•User-given aliases ("Tasks DB", "Main workspace")
•Commonly used view IDs
•Parent-child relationships (which databases are under which pages)
•Any user preference expressed during interaction

What NOT to Store

Never store token_v2, credentials, API keys, or any sensitive data. Never store full page content (just IDs and titles). Never store block-level IDs unless they're persistent references (like database blocks).

Handling Stale Data

If a memorized ID returns an error (page not found, access denied), remove it from MEMORY.md. Don't blindly trust memorized data — verify when something seems off. Prefer re-searching over using a memorized ID that might be stale.

Format / Example

Here's a concrete example of how to structure your MEMORY.md:

markdown

# Agent Notion Memory

## Workspaces

- `abc123-...` — Acme Corp (default)

## Pages (Acme Corp)

- `page-id-1` — Product Roadmap (top-level)
- `page-id-2` — Q1 Planning (under Product Roadmap)

## Databases (Acme Corp)

- `coll-id-1` — Tasks (under Product Roadmap, views: `view-1`)
- `coll-id-2` — Contacts (top-level)

## Aliases

- "roadmap" → `page-id-1` (Product Roadmap)
- "tasks" → `coll-id-1` (Tasks database)

## Notes

- User prefers --pretty output for search results
- Main workspace is "Acme Corp"

Memory lets you skip repeated search and workspace list calls. When you already know an ID from a previous session, use it directly.

Commands

Auth Commands

bash

agent-notion auth extract    # Extract token_v2 from Notion desktop app
agent-notion auth status     # Check authentication status
agent-notion auth logout     # Remove stored token_v2

Page Commands

bash

# List pages in a space (top-level only)
agent-notion page list --workspace-id <workspace_id> --pretty
agent-notion page list --workspace-id <workspace_id> --depth 2 --pretty

# Get a page and all its content blocks
agent-notion page get <page_id> --workspace-id <workspace_id> --pretty
agent-notion page get <page_id> --workspace-id <workspace_id> --limit 50
agent-notion page get <page_id> --workspace-id <workspace_id> --backlinks --pretty

# Create a new page under a parent
agent-notion page create --workspace-id <workspace_id> --parent <parent_id> --title "My Page" --pretty

# Create a page with markdown content
agent-notion page create --workspace-id <workspace_id> --parent <parent_id> --title "My Doc" --markdown '# Hello\n\nThis is **bold** text.'

# Create a page with markdown from a file
agent-notion page create --workspace-id <workspace_id> --parent <parent_id> --title "My Doc" --markdown-file ./content.md

# Replace all content on a page with new markdown
agent-notion page update <page_id> --workspace-id <workspace_id> --replace-content --markdown '# New Content'
agent-notion page update <page_id> --workspace-id <workspace_id> --replace-content --markdown-file ./updated.md

# Update page title or icon
agent-notion page update <page_id> --workspace-id <workspace_id> --title "New Title" --pretty
agent-notion page update <page_id> --workspace-id <workspace_id> --icon "🚀" --pretty

# Archive a page
agent-notion page archive <page_id> --workspace-id <workspace_id> --pretty

Database Commands

bash

# Get database schema
agent-notion database get <collection_id> --workspace-id <workspace_id> --pretty

# Query a database (auto-resolves default view)
agent-notion database query <collection_id> --workspace-id <workspace_id> --pretty
agent-notion database query <collection_id> --workspace-id <workspace_id> --limit 10 --pretty
agent-notion database query <collection_id> --workspace-id <workspace_id> --view-id <view_id> --pretty
agent-notion database query <collection_id> --workspace-id <workspace_id> --search-query "keyword" --pretty
agent-notion database query <collection_id> --workspace-id <workspace_id> --timezone "America/New_York" --pretty

# List all databases in workspace
agent-notion database list --workspace-id <workspace_id> --pretty

# Create a database
agent-notion database create --workspace-id <workspace_id> --parent <page_id> --title "Tasks" --pretty
agent-notion database create --workspace-id <workspace_id> --parent <page_id> --title "Tasks" --properties '{"status":{"name":"Status","type":"select"}}' --pretty

# Update database title or schema
agent-notion database update <collection_id> --workspace-id <workspace_id> --title "New Name" --pretty

# Add a row to a database
agent-notion database add-row <collection_id> --workspace-id <workspace_id> --title "Row title" --pretty
agent-notion database add-row <collection_id> --workspace-id <workspace_id> --title "Row title" --properties '{"Status":"In Progress","Due":{"start":"2025-03-01"}}' --pretty

# Get view configuration and property visibility
agent-notion database view-get <view_id> --workspace-id <workspace_id> --pretty

# Show or hide properties on a view (comma-separated names)
agent-notion database view-update <view_id> --workspace-id <workspace_id> --show "ID,Due" --pretty
agent-notion database view-update <view_id> --workspace-id <workspace_id> --hide "Assignee" --pretty
agent-notion database view-update <view_id> --workspace-id <workspace_id> --show "Status" --hide "Due" --pretty

Block Commands

bash

# Get a specific block
agent-notion block get <block_id> --workspace-id <workspace_id> --pretty
agent-notion block get <block_id> --workspace-id <workspace_id> --backlinks --pretty

# List child blocks
agent-notion block children <block_id> --workspace-id <workspace_id> --pretty
agent-notion block children <block_id> --workspace-id <workspace_id> --limit 50 --pretty
agent-notion block children <block_id> --workspace-id <workspace_id> --start-cursor '<next_cursor_json>' --pretty

# Append child blocks
agent-notion block append <parent_id> --workspace-id <workspace_id> --content '[{"type":"text","properties":{"title":[["Hello world"]]}}]' --pretty

# Append markdown content as blocks
agent-notion block append <parent_id> --workspace-id <workspace_id> --markdown '# Hello\n\nThis is **bold** text.'

# Append markdown from a file
agent-notion block append <parent_id> --workspace-id <workspace_id> --markdown-file ./content.md

# Update a block
agent-notion block update <block_id> --workspace-id <workspace_id> --content '{"properties":{"title":[["Updated text"]]}}' --pretty

# Delete a block
agent-notion block delete <block_id> --workspace-id <workspace_id> --pretty

Block Types Reference

The internal API uses a specific block format. Here are all supported types:

Headings

json

{"type": "header", "properties": {"title": [["Heading 1"]]}}
{"type": "sub_header", "properties": {"title": [["Heading 2"]]}}
{"type": "sub_sub_header", "properties": {"title": [["Heading 3"]]}}

Text

json

{"type": "text", "properties": {"title": [["Plain text paragraph"]]}}

Lists

json

{"type": "bulleted_list", "properties": {"title": [["Bullet item"]]}}
{"type": "numbered_list", "properties": {"title": [["Numbered item"]]}}

To-Do / Checkbox

json

{"type": "to_do", "properties": {"title": [["Task item"]], "checked": [["Yes"]]}}
{"type": "to_do", "properties": {"title": [["Unchecked task"]], "checked": [["No"]]}}

Code Block

json

{"type": "code", "properties": {"title": [["console.log('hello')"]], "language": [["javascript"]]}}

Quote

json

{"type": "quote", "properties": {"title": [["Quoted text"]]}}

Divider

json

{"type": "divider"}

Rich Text Formatting

Rich text uses nested arrays with formatting codes:

Format	Syntax	Example
Plain	`[["text"]]`	`[["Hello"]]`
Bold	`["text", [["b"]]]`	`["Hello", [["b"]]]`
Italic	`["text", [["i"]]]`	`["Hello", [["i"]]]`
Strikethrough	`["text", [["s"]]]`	`["Hello", [["s"]]]`
Inline code	`["text", [["c"]]]`	`["Hello", [["c"]]]`
Link	`["text", [["a", "url"]]]`	`["Click", [["a", "https://example.com"]]]`
Bold + Italic	`["text", [["b"], ["i"]]]`	`["Hello", [["b"], ["i"]]]`

Multiple segments: [["plain "], ["bold", [["b"]]], [" more plain"]]

Comment Commands

bash

# List comments on a page
agent-notion comment list --page <page_id> --workspace-id <workspace_id> --pretty

# Create a comment on a page (starts a new discussion)
agent-notion comment create "This is a comment" --page <page_id> --workspace-id <workspace_id> --pretty

# Reply to an existing discussion thread
agent-notion comment create "Replying to thread" --discussion <discussion_id> --workspace-id <workspace_id> --pretty

# Get a specific comment by ID
agent-notion comment get <comment_id> --workspace-id <workspace_id> --pretty

Search Command

bash

# Search across workspace (--workspace-id is required)
agent-notion search "query" --workspace-id <workspace_id> --pretty
agent-notion search "query" --workspace-id <workspace_id> --limit 10 --pretty
agent-notion search "query" --workspace-id <workspace_id> --start-cursor <offset> --pretty

User Commands

bash

# Get current user info
agent-notion user me --pretty

# Get a specific user
agent-notion user get <user_id> --workspace-id <workspace_id> --pretty

Output Format

JSON (Default)

All commands output JSON by default for AI consumption:

bash

# Search results
agent-notion search "Roadmap" --workspace-id <workspace_id>

json

{
  "results": [
    {
      "id": "305c0fcf-90b3-807a-bc1a-dc7cc18e0022",
      "title": "Getting Started",
      "score": 76.58
    }
  ],
  "has_more": true,
  "next_cursor": "20",
  "total": 100
}

bash

# Database query — properties use human-readable field names from the collection schema
agent-notion database query <collection_id> --workspace-id <workspace_id>

json

{
  "results": [
    {
      "id": "row-uuid",
      "properties": {
        "Name": "Acme Corp",
        "Status": "Active",
        "Type": "Enterprise"
      }
    }
  ],
  "has_more": false,
  "next_cursor": null
}

bash

# Page get — returns page metadata with content blocks
agent-notion page get <page_id> --workspace-id <workspace_id>

json

{
  "id": "page-uuid",
  "title": "My Page",
  "blocks": [
    { "id": "block-1", "type": "text", "text": "Hello world" },
    { "id": "block-2", "type": "to_do", "text": "Task item" }
  ]
}

bash

# With --backlinks: includes pages that link to this page/block
agent-notion page get <page_id> --workspace-id <workspace_id> --backlinks
agent-notion block get <block_id> --workspace-id <workspace_id> --backlinks

json

{
  "id": "page-uuid",
  "title": "My Page",
  "blocks": [...],
  "backlinks": [
    { "id": "linking-page-uuid", "title": "Page That Links Here" }
  ]
}

bash

# Block get — collection_view blocks include collection_id and view_ids
agent-notion block get <block_id> --workspace-id <workspace_id>

json

{
  "id": "block-uuid",
  "type": "collection_view",
  "text": "",
  "parent_id": "parent-uuid",
  "collection_id": "collection-uuid",
  "view_ids": ["view-uuid"]
}

Pretty (Human-Readable)

Use --pretty flag for formatted output on any command:

bash

agent-notion search "Roadmap" --workspace-id <workspace_id> --pretty

When to Use `--backlinks`

Backlinks reveal which pages/databases link to a given page. This is critical for efficient navigation.

Use --backlinks when:

•Tracing relations: A search result looks like a select option, enum value, or relation target (e.g., a plan name or category). Backlinks instantly reveal all rows/pages that reference it via relation properties — no need to hunt for the parent database.
•Finding references: You found a page and want to know what other pages mention or link to it.
•Reverse lookups: Instead of querying every database to find rows pointing to a page, use backlinks on the target page to get them directly.

Example — finding who uses a specific plan:

bash

# BAD: 15 API calls — search, open empty pages, trace parents, find database, query
agent-notion search "Enterprise Plan" ...
agent-notion page get <plan-page-id> ...  # empty
agent-notion block get <plan-page-id> ...  # find parent
# ... many more calls to discover the database

# GOOD: 2-3 API calls — search, then backlinks on the target
agent-notion search "Enterprise Plan" ...
agent-notion page get <plan-page-id> --backlinks --pretty
# → backlinks immediately show all people/rows linked to this plan

Pagination

Commands that return lists support pagination via has_more, next_cursor fields:

•block children: Cursor-based. Pass next_cursor value from previous response as --start-cursor.
•search: Offset-based. Pass next_cursor value (a number) as --start-cursor.
•database query: Use --limit to control page size. has_more indicates more results exist, but the private API does not support cursor-based pagination — increase --limit to fetch more rows.

Limitations

•auth extract supports macOS and Linux. Windows DPAPI decryption is not yet supported.
•token_v2 uses the unofficial internal API and may break if Notion changes it.
•This is a private/unofficial API and is not supported by Notion.

Agent Notion

Quick Start

Authentication

Token Extraction (Desktop App)

Memory

Reading Memory

Writing Memory

What to Store

What NOT to Store

Handling Stale Data

Format / Example

Commands

Auth Commands

Page Commands

Database Commands

Block Commands

Block Types Reference

Headings

Text

Lists

To-Do / Checkbox

Code Block

Quote

Divider

Rich Text Formatting

Comment Commands

Search Command

User Commands

Output Format

JSON (Default)

Pretty (Human-Readable)

When to Use --backlinks

Pagination

Limitations

When to Use `--backlinks`