
ArticleJiraConfluence
Atlassian MCP server permissions: what ChatGPT, Claude and Cursor can read and write
Org admins get two places to limit what external AI tools do through the Atlassian MCP server: the Permissions tab and a Guard data security policy. We tested both on our own org. Blocking the Jira write toolset stopped Jira writes from OAuth and API-token connections within 18 seconds. The data security policy only held on the V2 endpoint.