The gist
Notion AI turns a static knowledge base into a living team assistant: you write through Claude, Notion stores it, your team reads and edits together, and your agents know what has already been researched.
Key concepts
- Notion AI is a built-in AI assistant on top of Notion, separate from Claude Code
- Notion MCP is the bridge between Claude Code and your Notion knowledge base (as of October 2026, the main option is Notion's hosted server with OAuth sign-in)
- Workflow: Claude writes → Notion stores → the team (or other agents) reads
- Agents use Notion as "memory" so they don't research the same thing twice
- AI page autofill: Notion suggests content on its own based on the page structure
Theory
What Notion AI is and how it differs from regular Notion
Regular Notion is a very flexible notebook: tables, pages, databases, kanban boards. Powerful, but passive: you put things in, you take things out.
Notion AI adds a layer of intelligence right inside the interface:
- Auto-summaries of blocks and pages
- Autocomplete for text right in the editor
- Search by meaning (not just by keywords)
- Content generation from a page template
- Answers to questions about the team's knowledge base
The pricing question (as of October 2026):
- Full Notion AI (Notion Agent for multi-step tasks on your data, plus AI Meeting Notes) is included in the Business plan
- On Free and Plus, AI works as a trial
- Custom Agents (autonomous agents inside Notion) use credits, with a free trial first
- If you work solo: start with the free plan and the AI trial, and move to a paid plan when the trial isn't enough
- Current prices: notion.com/pricing; for a general overview of prices, see What's current
Notion MCP: the bridge between Claude Code and Notion
On its own, Claude Code doesn't know what's in your Notion. Notion MCP fixes that: it gives Claude Code tools to search, read, create and edit pages.
Setup (the main way). Notion runs its own hosted MCP server. You don't need to install anything on your computer; you sign in with OAuth:
# Add the hosted Notion MCP to Claude Code
claude mcp add --transport http notion https://mcp.notion.com/mcpThen, inside Claude Code, type /mcp and complete the OAuth sign-in: choose your workspace and allow access. The --scope project flag saves the connection in the project's .mcp.json so the whole team can use it.
Backup option: your own integration token. The old local server @notionhq/notion-mcp-server is no longer being developed as of October 2026, and Notion recommends the hosted version. You only need a token if OAuth isn't available. The .mcp.json config (check the current API version in Notion's documentation):
{
"mcpServers": {
"notion": {
"command": "npx",
"args": ["-y", "@notionhq/notion-mcp-server"],
"env": {
"OPENAPI_MCP_HEADERS": "{\"Authorization\": \"Bearer ntn_YOUR_TOKEN\", \"Notion-Version\": \"2022-06-28\"}"
}
}
}
}How to get a token (backup option only):
- Go to notion.com/my-integrations
- Create an integration → copy the Internal Integration Token
- Share the databases you need with the integration (⋯ → Connections → Add)
Once it's set up, Claude Code sees Notion as part of its toolkit. Give write access only to the pages and databases the agent actually needs to work with: anything it can access, it can change.
Workflow: Claude writes → Notion stores → the team reads
The main pattern for working with Notion MCP:
Claude Code does the task
↓
The results (research, code, decisions) are saved to Notion
↓
The team sees them in real time
↓
The next agents read Notion before they start workingExample prompt for Claude:
Research the competitors in niche X. Save the results to Notion in the "Competitive Intelligence" database: create a separate page for each competitor with these fields: name, website, pricing, strengths, weaknesses.
Claude does the research and writes it straight into Notion in a structured way, with no manual copying.
Notion as agent memory
One of the most powerful ideas: agents can read Notion before they start working, so they don't redo work that's already been done.
The "agent memory" pattern:
# At the start of every research agent:
# 1. Check Notion: has this already been researched?
# 2. If yes, use what's there
# 3. If not, do the research and write it to Notion
system_prompt = """
Before starting research, query the Notion database "Research Log".
If the topic has already been researched (there is a page with a similar request),
return the existing data instead of doing new research.
If the topic isn't there, do the research and create a new page.
"""This saves money (fewer API requests) and time (no duplicated work).
AI page autofill in Notion
Notion AI can fill in pages right in the interface:
- Select text → open AI from the menu → get an explanation, a summary or a translation
- Start a page → ask AI to continue → it keeps going in the same style
- Empty database → ask AI to generate → it suggests a structure for the rows
The button names in Notion's interface change, so look for the AI option in the block menu and the page toolbar.
Especially useful for databases: In tables, Notion AI can fill in a field automatically based on other fields. For example: you have a "Product description" column → AI automatically fills in the "Tags" and "Target audience" columns.
Three ready-made templates to start with
1. Team knowledge base
Database: Knowledge Base Fields: Title | Category | Tags | Created | Author | Status Categories: Processes / Product / Tech / Competitors / Clients AI integration: Claude creates pages automatically after each research task
2. CRM in Notion
Database: Clients Fields: Name | Company | Status | Source | Next step | Notes Statuses: Lead / Negotiating / Client / Closed AI integration: Claude adds records after calls, fills in "Next step" automatically
3. Content calendar
Database: Content Calendar Fields: Title | Format | Platform | Date | Status | Brief | Final text Formats: Post / Video / Email / Article AI integration: Claude generates the Brief and Final text from the title
Practice
Connect Notion MCP with
claude mcp add --transport http notion https://mcp.notion.com/mcpand sign in through/mcp(the token-based backup option is in the section above)Create a test page in Notion called "AI Research Log"
If you're connecting with a token, share this page with your integration (⋯ → Connections). With OAuth sign-in, access is determined by your permissions in the workspace
Start Claude Code and check the connection:
# In Claude Code, ask:
"Show me the list of pages in my Notion workspace"- Ask Claude to create a page:
Create a page in Notion titled "MCP test [date]". Add a short description of what Notion MCP is, three usage examples and a link to the Notion API documentation.
Create a "Content Calendar" database from the template above (if you connected with a token, give the integration access)
Ask Claude to fill in 3 rows of the content calendar for next week for your project
Tools and resources
- Notion: the platform itself (free plan to start)
- Notion MCP: the hosted MCP server from the Notion team (the main way)
- notion-mcp-server: the older local server (no longer being developed)
- Notion API Docs: documentation for custom integrations
- Notion AI: the built-in AI assistant
- Prices and versions: What's current
Key takeaways
Notion MCP turns Claude Code from a one-person tool into a team system: what your agents produce lives in a shared knowledge base.
Agent memory in Notion solves the duplication problem: agents first read what's already been done and only then get to work, like a good employee who checks the archive before starting new research.
Three templates (knowledge base, CRM, content calendar) are the minimum set for an AI entrepreneur who wants to stop keeping everything in their head.
What's next
→ The philosophy of folders: the PARA system for AI entrepreneurs: how to organize your knowledge and projects, wherever they're stored
The mark stays in this browser only and is never sent anywhere. My progress