notion-private-api-mcp
<!-- mcp-name: io.github.kirvigen/notion-private-api-mcp -->
# Notion Private API MCP Server
[](LICENSE)
[](https://nodejs.org/)
[](https://modelcontextprotocol.io/)
[](#contributing)
> **An unofficial Notion MCP server built on Notion's private API (`token_v2`).** It gives
> Claude Desktop, Claude Code, Cursor and any other MCP client **full read/write access to your
> entire Notion workspace — with no integration token and without sharing pages one by one.**
Unlike servers built on the official Notion API, this one authenticates with your browser
session cookie, so an LLM agent can search, read, create and edit **any** page your account can
see — instantly, with zero setup in Notion. Built on the official
[`@modelcontextprotocol/sdk`](https://www.npmjs.com/package/@modelcontextprotocol/sdk) (stdio transport).
---
## Why this server?
| | **This server** (private API) | Official Notion API / MCP |
|---|---|---|
| Setup in Notion | None — just your browser cookie | Create an integration + share each page |
| Access scope | **Everything your account can see** | Only pages explicitly shared with the integration |
| Auth | `token_v2` session cookie | Integration token / OAuth |
| Best for | Personal automation, full-workspace agents | Production apps, multi-user, official support |
| Stability | ⚠️ Fragile, undocumented | ✅ Stable, supported |
If you just want an agent over **your own** workspace without fighting integration permissions,
this is the fastest path. For production / multi-tenant apps, use the
[official Notion MCP server](https://github.com/makenotion/notion-mcp-server).
---
## Table of contents
- [Important: private API](#️-important-this-uses-notions-private-api)
- [Quick start](#quick-start)
- [Tools](#tools)
- [Configuration](#configuration)
- [Usage with MCP clients](#usage-with-mcp-clients)
- [Example prompts](#example-prompts)
- [Writing content](#writing-content)
- [Troubleshooting / FAQ](#troubleshooting--faq)
- [Contributing](#contributing)
- [License](#license) · [Disclaimer](#disclaimer)
---
## ⚠️ Important: this uses Notion's private API
This server talks to Notion's **undocumented internal API** (`https://www.notion.so/api/v3`),
**not** the official public API:
- Auth is your browser cookie (`token_v2`) — effectively your account password. **Never commit it.**
- Notion can change or break this API at any time, and using it may be against Notion's ToS.
- It is inherently fragile and **not for production** — use at your own risk, with your own data.
---
## Quick start
```bash
git clone https://github.com/kirvigen/notion-private-api-mcp.git
cd notion-private-api-mcp
npm install
export NOTION_TOKEN_V2='your_token_v2' # see "Configuration" below
npm start
```
Then register it in your MCP client ([Claude Desktop / Claude Code](#usage-with-mcp-clients)).
---
## Tools
| Tool | Description |
|---|---|
| `get_page` | Read a page block and its metadata |
| `get_block` | Read a single block |
| `get_block_children` | Read the direct child blocks of a page or block |
| `get_style_documentation` | Catalog of supported block types & inline annotations (call before composing complex pages) |
| `markdown_to_blocks` | Preview how Markdown parses into the simplified block JSON |
| `create_page` | Create a child page under another page, from blocks or Markdown |
| `append_blocks` | Append blocks/Markdown to a page (at the end, or after a given block) |
| `replace_page_content` | Replace the direct child blocks of a page |
| `update_block_text` | Replace the plain-text content of a block (e.g. a code block) |
| `delete_blocks` | Remove (archive) direct child blocks from a page |
| `sync_markdown_file` | Create or replace a page from a local Markdown file |
> Tool parameters are defined in [`src/server.js`](src/server.js).
---
## Configuration
Configured entirely through environment variables:
| Variable | Required | Description |
|---|---|---|
| `NOTION_TOKEN_V2` | ✅ | Your Notion session cookie (`token_v2`) |
| `NOTION_PRIVATE_API_BASE` | — | API base URL (default: `https://www.notion.so`) |
### Getting your `token_v2`
1. Log in to Notion in your browser: <https://www.notion.so>
2. Open DevTools (F12) → **Application** → **Cookies** → `https://www.notion.so`
3. Copy the value of the `token_v2` cookie.
> 🔒 Treat it like a password. Keep it in your shell env or an untracked `.env` — never commit it.
```bash
cp .env.example .env # then edit .env
```
---
## Usage with MCP clients
### Claude Desktop
Add to `claude_desktop_config.json`:
```json
{
"mcpServers": {
"notion-private": {
"command": "node",
"args": ["/absolute/path/to/notion-private-api-mcp/src/server.js"],
"env": { "NOTION_TOKEN_V2": "your_token_v2" }
}
}
}
```
### Claude Code
```bash
claude mcp add notion-private \
--scope local \
--env NOTION_TOKEN_V2='your_token_v2' \
-- node /absolute/path/to/notion-private-api-mcp/src/server.js
```
The helper scripts `./run-desktop.sh` and `./run-codex.sh` resolve the repo path automatically
and log to `/tmp`.
---
## Example prompts
Once connected, just talk to your agent:
- *"Find my 'Q3 Roadmap' page in Notion and summarize it."*
- *"Create a child page under <page> titled 'Meeting notes' with today's action items."*
- *"Append a TODO list to <page> with these three tasks…"*
- *"Sync my local `CHANGELOG.md` into the release-notes page."*
---
## Writing content
Tools that write accept a plain-JSON **simplified block** format:
```json
[
{ "type": "heading_1", "text": "Release Notes" },
{ "type": "paragraph", "text": "First paragraph." },
{ "type": "to_do", "text": "Ship the feature", "checked": true },
{ "type": "toggle", "text": "Details", "children": [
{ "type": "bulleted_list_item", "text": "Item one" }
]}
]
```
Supported types: `paragraph`, `heading_1`/`2`/`3`, `bulleted_list_item`, `numbered_list_item`,
`to_do`, `toggle`, `quote`, `callout`, `code`, `divider`.
You can also pass **Markdown** (a stable subset: headings, paragraphs, bullet/numbered lists,
task items, blockquotes, fenced code, horizontal rules). Call `get_style_documentation` from your
client for the authoritative, machine-readable catalog. Nested lists, tables and inline formatting
are not implemented yet.
---
## Troubleshooting / FAQ
**Where do I get `token_v2`?** See [Getting your token_v2](#getting-your-token_v2).
**`NOTION_TOKEN_V2 is required`** — the env var isn't set in the process that launches the server
(check your MCP client's `env` block, not just your shell).
**My token stopped working** — `token_v2` expires when your Notion session ends (logout, password
change, long inactivity). Grab a fresh cookie and update it.
**`MemcachedCrossCellError`** — a transient Notion routing error on multi-cell workspaces. The
client already retries and falls back to `loadPageChunk`; just retry the call if it surfaces.
**Is this against Notion's ToS?** It uses an undocumented internal API. Use only with your own
account and data, at your own risk.
---
## Project layout
```
src/
├── server.js # MCP server: tool registration + stdio transport
├── notion-client.js # Private-API client (cookie auth, transactions, retries)
├── notion-blocks.js # Builds Notion block trees from simplified blocks
├── markdown.js # Markdown → simplified-block parser
└── style-docs.js # Catalog returned by get_style_documentation
```
---
## Contributing
Issues and PRs are welcome. If this saved you time, please ⭐ the repo — it genuinely helps others
discover it.
---
## License
[MIT](LICENSE) © kir.vigen
## Disclaimer
Not affiliated with Notion Labs, Inc. Relies on an undocumented internal API; by using it you
accept all risks, including possible account restrictions and breakage when the API changes.
Use only with your own data.
TDQS
Scored across 11 tools
Each tool targets a distinct operation: page vs block CRUD, content replacement, style documentation, and markdown conversion. No meaningful overlap exists between any two tools.
Most tools follow a verb_noun pattern (e.g., get_page, delete_blocks). However, markdown_to_blocks deviates by using a descriptor format, breaking the uniform convention.
With 11 tools, the set covers core Notion page and block operations, plus supporting utilities. The count is well-scoped for the private API's capabilities without being excessive.
The set covers basic content creation, reading, and modification, but lacks page metadata updates, page deletion, and search. These gaps may hinder some workflows but core content operations are present.