How We Connected Notion Ai To 160 New Apps In Three Weeks
Source
notion.com
Author
Notion Blog
Date
Why it matters
Shows how to turn repetitive MCP connector onboarding into an agent-run kanban pipeline with humans on the key steps, and which MCP auth methods (DCR/CIMD vs bring-your-own credentials) cost the most effort.
Key takeaways · AI-distilled
Before the AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition →, Notion says each MCPThe Model Context Protocol — an open standard that lets any AI assistant plug into any tool or data source without custom integration code.Full definition → connection meant hours of sign-ups, CAPTCHAs and at least two rounds of testing; the easiest took about half an hour and harder ones took days, mostly spent waiting.
Anyone inside or outside Notion can request a new MCP, and a weekly flow adds servers based on the most popular custom MCPs. Each server is a card in a Notion database whose kanban columns are pipeline steps.
Verification happens twice: the MCP is added to the dev environment, Notion AI is asked to use the tool to confirm real data comes back, then the same checks run again in production.
Notion gave the agent its own browser (agent-browser) for sign-ups and settings, and it checks hourly whether async steps, like a logo deploy or waiting for business hours, let a card move forward.
Notion says people still review every code change, design logos, hand-test the most popular MCP servers and handle partner conversations.
Terms in this piece · Glossary
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
MCP — The Model Context Protocol — an open standard that lets any AI assistant plug into any tool or data source without custom integration code.
