> ## Content Index
> Fetch the complete content index at: https://www.vizney.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# Claude Dashboards Data Security: What Reaches Anthropic vs. Chat and MCP
- URL: https://www.vizney.com/claude-dashboards-data-security/
- Published: 2026-10-23T13:34:11.000Z
- Updated: 2026-10-23T13:34:11.000Z
- Author: Yehwan Hwang
- Tags: Knowledge Base

Claude Dashboards, released in beta in October 2026, stores the query results behind each chart on Anthropic’s servers along with the dashboard. With live queries, the data goes through Anthropic’s cloud every time the dashboard opens. Building with Claude in chat and connecting a database through MCP also send data to Anthropic, but the data leaves at a different point. This post breaks down what reaches Anthropic in each approach and what to check before a security review.

With chat-based development and MCP, data leaves only while you build; the finished dashboard queries the database directly from an internal server. A Claude dashboard, by contrast, lives on Anthropic’s servers. Data therefore passes through Anthropic every time someone opens it, long after the build is done. Here is how that plays out for a monthly sales dashboard on an on-premises Oracle database.

![Diagram comparing when data reaches Anthropic in three approaches, split into while you build and when someone opens the dashboard. Chat-based development sends table definitions, sample rows, and files Claude reads while you build; when someone opens the dashboard, data moves only between the internal server and Oracle. Database MCP sends every row queried during development, and viewer data stays inside your network. Claude Dashboards sends the query results for each chart while you build, and because the dashboard lives on Anthropic’s servers, data passes through Anthropic whenever someone opens it.](https://storage.ghost.io/c/6a/50/6a50f03c-1922-4e90-bf18-c5abd0f819d4/content/images/2026/10/claude-dashboards-data-timing-en-1.png)

- **Chat-based development**: You describe the tables and requirements, and Claude writes the SQL and front-end code. What reaches Anthropic is what you paste or attach (table definitions, sample rows, error messages) plus any files Claude reads. A `.env` file with database credentials also reaches Anthropic if Claude reads it, so keep credentials outside the paths Claude can read. Once the code is deployed to an internal server, viewer data travels only between that server and Oracle.
- **Database MCP**: With a local MCP server such as Oracle’s SQLcl MCP Server connected to Claude Desktop, queries run from your machine over the internal network. The results, however, return to the conversation as tool results and go to Claude. As a result, every row queried during development is sent to Anthropic with the conversation. After deployment, viewing works the same as with chat-based development.
- **Claude Dashboards**: Claude [writes and runs a SQL query for each chart](https://support.claude.com/en/articles/17454700-get-started-with-claude-dashboards?ref=vizney.com) and bundles the results into datasets. The dashboard is saved to your Claude account separately from the chat and can be shared. In [Tabular Editor’s testing](https://tabulareditor.com/blog/what-are-claude-dashboards?ref=vizney.com), a local MCP server could only supply extracts. The query results are then stored as files inside the dashboard, and anyone who can open the dashboard can read them. Live queries that refresh on open need a remote MCP server, and [remote MCP servers are reached from Anthropic’s cloud](https://support.claude.com/en/articles/11175166-get-started-with-custom-connectors-using-remote-mcp?ref=vizney.com). To connect an on-premises Oracle database live, you must open your firewall so that Anthropic’s IP ranges can reach the MCP server in front of it.

![Diagram of the extract and live-query paths for Claude Dashboards, split between the internal network and Anthropic’s cloud. With an extract, a local MCP server on your machine queries Oracle, and the query results are stored as files in the dashboard in Anthropic’s cloud. With a live query, the dashboard refreshes on open by connecting from Anthropic’s IP ranges to a remote MCP server inside the internal network, so you must open your firewall.](https://storage.ghost.io/c/6a/50/6a50f03c-1922-4e90-bf18-c5abd0f819d4/content/images/2026/10/claude-dashboards-extract-live-en.png)

### What to check before a security review

- **Plan and training settings**: On consumer plans (Free, Pro, Max), turning on the model improvement setting (Help improve Claude) allows training on your chats and [retention in de-identified form for up to five years](https://privacy.claude.com/en/articles/10023548-how-long-do-you-store-my-data?ref=vizney.com). Team, Enterprise, and API usage falls under the [Commercial Terms](https://www.anthropic.com/legal/commercial-terms?ref=vizney.com) and is not used for training by default. Training and transmission are separate questions, though. On every plan, chats and dashboards are stored on Anthropic’s servers.
- **Retention and deletion**: Deleted chats are generally [removed from backend storage within 30 days](https://privacy.claude.com/en/articles/7996866-how-long-do-you-store-my-organization-s-data?ref=vizney.com). Dashboards are saved separately in the Artifacts tab, so manage them separately from chats.
- **ZDR scope**: A [Zero Data Retention agreement](https://privacy.claude.com/en/articles/8956058-i-have-a-zero-data-retention-agreement-with-anthropic-what-products-does-it-apply-to?ref=vizney.com) covers only eligible APIs, products using your commercial organization’s API key, and Claude Code on Enterprise plans. It does not cover chats or dashboards in the Claude apps.
- **Feature access**: Claude Dashboards is in beta on Pro, Max, Team, and Enterprise. On Enterprise it is off by default, and owners can turn it on or limit it to specific groups.
- **Sharing**: [Dashboards that use connectors can be shared only within your organization](https://support.claude.com/en/articles/14729249-use-artifacts-in-claude-cowork?ref=vizney.com), and Enterprise plans keep external sharing off by default. Consumer plans offer two options: Only you and Anyone with the link. Because extracts are stored inside the dashboard, its sharing scope is its data exposure scope.
- **MCP path and credentials**: Local MCP servers connect over your machine’s network, while remote MCP servers are reached from Anthropic’s cloud. On Team and Enterprise, a connector that uses a shared credential such as an API key gives everyone who uses it that credential’s access.
- **Query scope**: The queries Claude runs determine how much data leaves. Give the MCP connection a dedicated account that can read only the views it needs, and leave PII columns out of those views.

Which approach to use depends on the data involved.

- **Exploring public or approved data**: Claude Dashboards is a good fit. One question produces a dashboard, and each chart shows its query.
- **Development on sensitive sources**: Use database MCP, with a dedicated account and views to limit what leaves.
- **Dashboards many people use over time**: Run them on your BI platform, or deploy what you built through chat or MCP to an internal server. Viewer data stays inside your network.

Claude Dashboards is in beta, so recheck sharing and connector behavior in the [official help center](https://support.claude.com/en/articles/17454700-get-started-with-claude-dashboards?ref=vizney.com) when you adopt it.