Notion: Context for your agents - Switch AI agents without re-explaining your work
by•
Notion 3.8 turns your workspace into the shared context layer for every AI agent your team uses. Unlike chat tools that start from zero each time you switch models, Notion keeps your docs, databases, and Skills in one place any MCP agent can read and update. Connect Claude, ChatGPT, or any MCP agent, plug Notion Agents into 150+ tools, sync Skills everywhere with the new Skills API, and see every agent edit in upgraded version history. Plus column permissions and rebuilt synced databases.
Replies
Hey Product Hunt! 👋
Every time I try a new AI agent, I end up pasting in the same docs and re-explaining how we work. The agent changes, the context doesn’t.
Notion 3.8 fixes that by making your workspace the shared context layer for any agent you use.
Connect any MCP-compatible agent to Notion ➡️
➡️ It reads your docs, databases, and tasks for context
➡️ It updates them directly, no copy-pasting
➡️ Every edit is attributed in version history and can be restored in a click
What's new -
Skills API: write a Skill once, sync it to ChatGPT, Claude, and more
150+ MCP connections for Notion Agents (HubSpot, Stripe, Cloudflare)
Column permissions: control what each person or agent can see or edit
Use your ChatGPT plan in Notion with no markup (Business and Enterprise)
How many AI agents are you juggling right now, and where does your context live?
@kate_ramakaieva My context lives in syndicates of secret gists. Its interesting because its secret for everyone unknown to the direct secret url. So very accessible to all agents. yet private enough to stay out of the lime light. How does notion solve the privacy? Does it require auth?
Very interesting... is this something like creating a second brain inside Notion, instead of Obsidian for example, and let other LLMs working on it? Just to understand... thanks.
Really can't imagine notion positions itself as a context container
Swapping between Claude and ChatGPT mid project has always meant pasting the same context all over again, so a shared layer that any MCP agent can actually read and update feels like a real relief.