Atlassian on Feb. 25 launched an open beta of agents in Jira, allowing teams to assign tasks, set deadlines and track progress for AI agents from the same interface used to manage human workers, the company said in a press release.
The release also brought the Rovo Model Context Protocol (MCP) Server to general availability — an Atlassian-hosted server that gives MCP-compatible AI clients a single authenticated connection to Jira and Confluence, per Atlassian’s blog.
Agents inside the permission layer
The governance design is the release’s most deliberate technical choice. Because agents in Jira operate inside Jira’s existing permission structures, project configurations and approval workflows, approved agent updates are captured alongside human work item history.
That design addresses the compliance objection that tends to slow enterprise agent deployments: when AI takes action inside a business process, security and legal teams require a traceable log.
“Without clear coordination, [work] can easily turn into chaos,” said Tamar Yehoshua, Atlassian’s chief product and AI officer.
Sanchan Saxena, SVP and GM of Atlassian’s Teamwork Collection, told Techzine the challenge for enterprises is no longer identifying agent use cases; it’s preventing agent activity from creating coordination overhead rather than reducing it.
“What we don’t want is agents that do 10 times the work but also create 10 times the chaos,” Saxena said.
What the usage numbers say
That operational concern has quantifiable weight in Atlassian’s own data. Nearly one-third of all agentic MCP operations generated by its customers are writes — actions that create or update records rather than read them — Atlassian told SiliconANGLE.
Enterprises account for nearly 50% of all Rovo MCP Server usage, and customers on paid Atlassian editions drive 93% of that activity, per the press release. Atlassian’s Support documentation recommends administrators review high-impact changes before confirming and monitor audit logs for unusual activity.
The integration layer at general availability
The Rovo MCP Server is now generally available, moving Atlassian’s MCP integration layer out of beta. All traffic runs over HTTPS using TLS 1.2 or later, with OAuth 2.1 handling authentication, ensuring agent actions respect each user’s existing access controls, per Atlassian’s Support documentation.
Compatible AI clients include Claude by Anthropic, Cursor, Google’s Gemini CLI, Lovable and WRITER, the company said. Atlassian also launched the Rovo MCP Gallery simultaneously, enabling Rovo agents to pull live context from and take action in third-party tools. Launch partners include Amplitude, Box, Canva, Figma and Intercom.
Saxena told Techzine that Jira is among the most frequently requested MCP connectors by third-party tool vendors — a function of Atlassian’s installed base across more than 350,000 customers and 80% of the Fortune 500.
How Atlassian’s model compares
The architectural decision behind Atlassian’s design differs from how competing platforms have built agent layers. Salesforce’s Agentforce agents act on CRM records and Data Cloud objects inside Salesforce’s data boundary, with the Atlas Reasoning Engine orchestrating those actions, as described in a Salesforce Developers blog post.
ServiceNow’s AI Agent Fabric, announced at Knowledge 2025, also uses MCP and Agent2Agent protocols but to connect agents outward from within the ServiceNow environment, with actions surfaced through the Workflow Data Fabric layer, according to a ServiceNow press release.
Atlassian wrote on its blog about how it inverts that direction: external AI clients connect inward to Jira via the Rovo MCP Server, and agent actions are recorded inside Jira work items alongside human activity.