---
title: How do you govern the Atlassian Remote MCP Server for Jira and Confluence?
url: https://opencontroller.lyzr.com/blog/govern-atlassian-remote-mcp-server
published: 2026-11-17
updated: 2026-11-17
author: Lyzr
---

# How do you govern the Atlassian Remote MCP Server for Jira and Confluence?

**The Atlassian MCP server, earlier called the Atlassian Remote MCP Server and the Rovo MCP server, lets AI tools such as Claude and ChatGPT act in Jira, Confluence and Compass with each user's existing Atlassian permissions. Admins govern it with three controls: which client domains may connect over OAuth 2.1, whether API token access is allowed at all, and organization IP allowlists. There is no per-tool write switch, so write risk is managed through user permissions and review.**

Connecting an AI assistant to Jira is one of the most common MCP requests in enterprises. It is also one of the easiest to get wrong, because the assistant inherits everything the user can do, including editing and creating issues and pages.

## How does access work?

| Element | What Atlassian documents |
|---|---|
| Products | Jira, Confluence and Compass, with references to Teamwork Graph and Rovo |
| Default authentication | OAuth 2.1 through a user consent screen |
| Alternative authentication | API tokens, for non-interactive or shared-account automation, controlled by an organization setting |
| Permission model | Access is scoped to the user's existing Atlassian permissions |
| Example clients | Claude.ai and ChatGPT; Atlassian-supported domains cover AI partners such as Anthropic and OpenAI |
| Cost | Enriched Teamwork Graph API and tool calls, including unified search and context tools, consume Rovo credits |

For a tool call to succeed, three conditions must hold: the client's OAuth 2.1 domain is allowed in the MCP server settings (where applicable), the request comes from an IP address allowed by any configured IP allowlist, and the authentication method is enabled for the organization.

## What can admins control?

1. **Domain allowlist for OAuth clients.** Admins can add and delete trusted domains. They can also deselect "Allow Atlassian supported domains" to block the supported list, but cannot block individual domains within it. Domain controls apply only to OAuth 2.1 tools.
2. **API token authentication.** Admins can turn API token access on or off. Domain lists do not govern API token tools, so this is the setting that closes that path.
3. **IP allowlists.** Managed in Atlassian Administration and applied to both OAuth and API token requests. Some AI tools call from their own outbound IP ranges, which you may need to add.

Atlassian's settings page does not describe a single on/off switch for the server or an admin control specific to write operations. Plan your controls around that.

## What policy should you set?

A practical baseline:

1. **Allow only the AI clients you have approved.** If your organization has approved one assistant, block the supported list and add only that client's domain.
2. **Turn API tokens off by default.** Turn them on only for a named automation, with a dedicated account that has the narrowest project and space permissions you can give it. Shared accounts hide who did what. See [API keys vs OAuth for AI agents](/blog/api-keys-vs-oauth-for-ai-agents).
3. **Use IP allowlists where you already have them,** and confirm the approved client's egress ranges are listed so users are not pushed toward unapproved workarounds.
4. **Review high-privilege users.** Because the assistant acts with the user's permissions, a Jira admin using an AI tool gives that tool admin reach. Consider separate admin accounts that are not used with AI clients.
5. **Keep a human check on writes.** Atlassian recommends reviewing high-impact changes before confirming. Most MCP clients show tool calls before running them; require users to keep confirmation on for create, edit and transition actions.
6. **Watch Rovo credit consumption** as an early signal of heavy or unexpected use.
7. **Monitor audit logs** for unusual bulk edits, page creation or permission changes after AI connections begin.

## How does this fit your MCP inventory?

Record the Atlassian MCP server as a remote, vendor-hosted MCP server in your inventory with owner, approved clients, auth methods allowed, and last review date. Our [MCP server inventory fields](/blog/mcp-server-inventory-fields) list covers the full record, and [official vs community MCP servers](/blog/official-vs-community-mcp-servers) explains why vendor-hosted servers still need review. If you also run Rovo agents, see [governing Atlassian Rovo agents](/blog/govern-atlassian-rovo-agents). The [MCP security best practices](https://modelcontextprotocol.io/specification/draft/basic/security_best_practices) in the protocol specification cover the token and consent risks behind these controls.

## How Opencontroller does this

Opencontroller keeps a governed registry of agents, versions, tools and MCP servers, and Opencontroller Edge discovers coding agents, skills and local MCP servers on developer devices, installed by IT via MDM. Delegated on-behalf-of identity and policy as code apply across the estate. See the [MCP governance guide](/blog/how-to-govern-mcp-servers), the [cross-platform governance guide](/blog/how-to-govern-ai-agents-across-platforms), the [AI agent control plane overview](/blog/what-is-an-ai-agent-control-plane), or [book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Does the Atlassian MCP server bypass Jira or Confluence permissions?**
No. Access is scoped to the connecting user's existing Atlassian permissions.

**Can admins block a specific AI client from the Atlassian MCP server?**
Admins can allow specific domains and can block the whole Atlassian-supported domain list, but cannot block individual domains within that list.

**Do domain allowlists cover API token connections?**
No. Domain controls apply only to OAuth 2.1 tools. Turn off API token authentication to close that path.

## Key facts

Atlassian MCP server (earlier Atlassian Remote MCP Server / Rovo MCP server): connects AI tools such as Claude.ai and ChatGPT to Jira, Confluence and Compass; access scoped to the user's existing permissions; OAuth 2.1 default, API tokens optional via org setting; admin controls: trusted domain allowlist (OAuth only), "Allow Atlassian supported domains" toggle (cannot block individual supported domains), API token on/off, IP allowlists in Atlassian Administration (apply to OAuth and API tokens). No per-tool write control documented. Enriched Teamwork Graph calls consume Rovo credits. Opencontroller: registry of agents, tools, MCP servers; Edge discovers local MCP servers on devices via MDM; on-behalf-of identity; policy as code. Pricing starts from $250K per deployment per year.

## Related

- [How do you govern Atlassian Rovo agents?](https://opencontroller.lyzr.com/blog/govern-atlassian-rovo-agents.md)
- [What should you record for every MCP server in your inventory?](https://opencontroller.lyzr.com/blog/mcp-server-inventory-fields.md)
- [Should you use official vendor MCP servers or community-built ones?](https://opencontroller.lyzr.com/blog/official-vs-community-mcp-servers.md)
- [More on Platform Guides](https://opencontroller.lyzr.com/blog/cluster/platform-guides)
