Atlassian announced three third-party AI agents running inside Confluence, its collaboration and knowledge management platform, using Model Context Protocol (MCP) as the integration layer. The agents, from Lovable, Replit, and Gamma, are rolling out in open beta starting next week alongside Remix, a new visual editor that transforms Confluence page content into charts, infographics, and diagrams.
What the Agents Do
Each agent operates directly from a Confluence page through Rovo Chat, carrying both content and organizational context into the partner tool:
Lovable converts product specs and ideas on a Confluence page into working prototypes. Replit turns technical documentation into starter applications that engineers can fork and extend. Gamma generates presentation decks from meeting notes and project pages.
“With Remix and agents in Confluence, a single page becomes the starting point for whatever comes next: a clear story for leaders, a prototype for builders, or a walkthrough for customers,” Sanchan Saxena, Atlassian’s SVP of teamwork collaboration, wrote in the announcement.
Admins enable partner agents through Atlassian Administration under Connected Apps. No custom scripting or agent creation is required, according to TechCrunch.
MCP as the Integration Protocol
The technical decision worth watching is the protocol choice. Atlassian built these integrations on MCP, and the company is positioning the protocol as open: any partner can build a Confluence-compatible agent by connecting an MCP server, without waiting for Atlassian to build a custom integration.
Context flows through what Atlassian calls the Teamwork Graph, a layer of work relationships built from over 100 billion data points across Atlassian products. When an agent carries content into Lovable or Replit, it includes authorship, project relationships, and connected decisions, not just raw text, per Atlassian’s blog post.
This is Atlassian’s second major agent integration in two months. In February, the company added AI agents to Jira, letting agents and humans work side-by-side in project management workflows.
The Workflow-Embedded Pattern
The broader trend is clear. Rather than building standalone agent platforms, enterprise software companies are embedding agents into the tools workers already use.
Salesforce launched Agentforce as a separate platform in 2024 but has since pushed AI capabilities into existing products, including turning Slackbot into an AI agent in January 2026, according to TechCrunch. OpenAI partnered with four major consulting firms through Frontier Alliances to embed its technology into clients’ existing stacks rather than selling standalone subscriptions.
Atlassian’s bet is that the wiki page, where knowledge already lives, is the right starting point for agent workflows. If that bet lands, MCP becomes the plumbing that connects enterprise knowledge bases to any agent-powered tool. The protocol is open, the first three partners are live, and every other SaaS vendor watching this rollout is now deciding whether to ship their own MCP server.