MCP tools & controls, Git directory sync, pivot tables, & more

MCP updates: better controls and more tools

New MCP tools for integrations

MCP-connected agents can now manage integrations end-to-end with three new tools: create_integration, attach_integration, and detach_integration. An agent can now create an integration and wire it into a project without a human step in between.

Better MCP controls for workspace admins

You can now see every MCP OAuth grant issued for your workspace and revoke any of them from Settings > Security > MCP OAuth connections.

A new workspace setting on the same page turns external MCP client access on or off for the whole workspace.

New block type: pivot tables

You can now summarize and cross-tabulate data with a dedicated pivot table block. Just drop in rows, columns, values, and an aggregation; no code required. It's a focused first version, and we'll expand it based on how people use it.

Git Directory Sync

A new Git integration flow that syncs each notebook to its own .deepnote file, so a project's notebooks show up as separate files in your repo instead of being bundled into a single multi-notebook .deepnote file. Existing projects on the old bundled format get a built-in migration path. More improvements are on the way.

Watch this walkthrough for a quick look at how the migration and sync work.

Collapsible sections in notebooks

Headings in notebooks are now collapsible, so you can fold away the parts you're not working on and move through a long notebook far more easily.

Better agent context

The AI agent now has access to our documentation, so questions about setup, configuration, and product behavior come back grounded in the actual docs.

Pick your AI model

The AI model selector in your settings now offers more fine-grained selection of the latest publicly available models from OpenAI or Anthropic. You can also try Anthropic’s latest Sonnet 5 in Deepnote!

SQL blocks in R environments

SQL blocks now work in R-based environments (Deepnote iR with libs). Query any Deepnote integration from a SQL block, and results land in an R dataframe, the R equivalent of the pandas dataframe you'd get in a Python notebook. Downstream blocks (including DataFrame SQL) can consume it from there.

BigQuery OAuth in the VS Code extension

The Deepnote VS Code extension (v1.5.0) now supports OAuth-based BigQuery integrations, so each user signs in with their own Google account instead of sharing a service account. Short-lived tokens, MFA, and least-privilege access apply automatically.

Because the OAuth flow is proxied through Deepnote, the same integration config and redirect URLs work in both places: if you already use BigQuery OAuth in Deepnote Cloud, the integration carries over.

Install the latest version from the VS Marketplace or Open VSX.

Setup details in the docs.