TL;DR:
Choose the official Atlassian MCP server for a hosted Confluence Cloud connection, Composio Connect for managed access across Confluence and other apps, or the self-managed
mcp-atlassianproject for Server/Data Center deployments.Atlassian MCP v2 supports reading, creating, updating, and searching Confluence content within approved permissions and scopes.
ChatGPT requires an eligible workspace with custom MCP connections enabled. Claude Desktop can connect to Atlassian’s hosted MCP server as a custom connector.
Review your organization's access policies and confirm the required permissions before enabling write actions.
Every time you need Confluence content in Claude or ChatGPT, you copy the page, paste it into the chat, and repeat the next time the page changes. An MCP connection removes that loop. It lets a compatible assistant search pages, pull documentation, and create or edit content, but only within the spaces and pages the connected account can already access.
The main choice is between Atlassian's hosted server (Cloud only), the self-managed open-source project (Cloud and Data Center), and Composio's managed layer (Cloud, with access to other apps alongside Confluence).
What is MCP and why use it for Confluence?
The Model Context Protocol (MCP) is a shared protocol that lets AI assistants connect to outside tools and data. Think of it as a universal connector: instead of pasting Confluence content into every chat window, the assistant fetches it directly. A Confluence MCP connection can let an assistant search pages, retrieve documentation, and perform permitted edits without manually pasting page content into every conversation.
There are three main approaches:
Official Atlassian MCP server: Atlassian-hosted, available for Atlassian Cloud, with OAuth 2.1 and optional API-token authentication.
Open-source
mcp-atlassian: A self-managed community project supporting Confluence Cloud and Server/Data Center. Confirm compatibility with your particular deployment before you deploy it.Composio Connect: A managed MCP connection that provides authorization workflows and access to Confluence alongside other applications.
Method 1: Connect Confluence to Claude using Atlassian's official MCP server
Atlassian's official server uses https://mcp.atlassian.com/v2/mcp. It works with supported MCP clients, including Claude, and requires an Atlassian Cloud site. Your organization's settings may restrict which clients can connect. Follow Atlassian's published setup guide for the current steps.
Prerequisites
Access to a Confluence Cloud site and the pages or spaces you want the assistant to use.
A compatible Claude client.
Permission to authorize the connection, or approval from your organization's Atlassian administrator if its policies restrict MCP access.
Step-by-step setup
Check admin settings. To manage client domains and authentication, select your organization in Atlassian Administration and open Connectors → Atlassian MCP server. The same guide identifies OAuth 2.1 as the default and explains how organization admins can enable API-token authentication. Use the Permissions tab to control read, write, and search access.
**Add a custom connector in Claude Desktop.**Open Customize → Connectors or look for connectors under Pluginsin the newer layout. Follow Claude’s custom connector instructions to add a web connector.
Enter the server URL. Name the connector Atlassian and enter
https://mcp.atlassian.com/v2/mcp, the endpoint in Atlassian’s MCP setup guide.Sign in and grant access. Complete the Atlassian sign-in flow and review the requested permissions.
Test a read action first. Ask Claude to summarize a page you can already access in Confluence. Test page creation only after your administrator has approved write access.
What you can do with the official server
Atlassian MCP v2 provides tools organized into read, write, and search permission groups. Its core Confluence actions include:
Read:
getConfluenceContentretrieves pages, blog posts, and other supported content.listConfluenceContentandlistConfluenceSpaceshelp browse content and spaces.Write:
createConfluenceContentcreates pages and other supported content types.updateConfluenceContentupdates existing content.Search:
searchConfluencesearches Confluence content using Confluence Query Language (CQL).
The server also supports operations such as attachment retrieval, comments, version history, and content restrictions. Not every tool appears immediately in every client: MCP v2 exposes primary tools directly and can discover additional tools as needed. For the current list, see Atlassian's supported tools.
The connection operates within the user's existing Confluence permissions and the allowed MCP permission groups. It does not confer access to spaces the user cannot access.
Method 2: Connect Confluence to Claude using Composio
Connecting an assistant to Confluence and other apps can leave your team managing separate OAuth flows and hosting a server for each integration. Composio Connect handles app authorization through a managed MCP endpoint, so you can connect supported apps without building and hosting a separate MCP server for each one. See Composio Connect documentation and the Confluence integration reference.
Step-by-step setup
Open Composio Connect. Follow the setup instructions for your MCP-compatible client in the Composio documentation.
Add the connection. Composio's shared MCP endpoint is
https://connect.composio.dev/mcp. In Claude, click the**+** button in the message box, then go to Connectors > Add connector > Add custom connector and enter the endpoint. If your client needs header-based authentication, use the consumer API key from Composio's Connect settings and send it in thex-consumer-api-keyheader. Use the consumer API key for this connection. A general-purpose developer project API key won't work here. See Composio's authentication guidance.Authorize Confluence. Ask the connected assistant to use a Confluence action. When prompted, open the authorization link and approve access to the intended Atlassian account.
Verify access. Test a read action before authorizing page creation or edits.
Composio's published Confluence integration includes actions for page creation, updates, searches, and other content operations. The Confluence integration reference lists the full set. Its managed connection handles app authorization and credential refresh, so your team does not need to implement those OAuth flows for each supported app. For example, connect Confluence today and add Slack or GitHub later without writing new auth code. According to Composio’s pricing page, Standard includes Firewall and single sign-on (SSO). Enterprise adds audit logs, SIEM streaming, and SCIM provisioning.
Method 3: Connect Confluence to ChatGPT
ChatGPT supports custom remote MCP connections in eligible workspaces, so connecting Confluence is not limited to Deep Research or custom GPT OpenAPI actions. Atlassian also lists ChatGPT among supported MCP clients. See the Atlassian setup guide and OpenAI's custom MCP connection documentation.
Option A: Connect the official Atlassian MCP server
In a ChatGPT workspace with custom MCP connections enabled, an authorized administrator or developer can configure https://mcp.atlassian.com/v2/mcp, complete the Atlassian authorization flow, review available actions, and publish the connection to permitted users. Availability and approval controls depend on your ChatGPT plan and workspace settings. Read and write permissions should be reviewed separately before deployment.
Option B: Connect through Composio
Composio Connect can expose authorized Confluence actions to a compatible MCP client. Use the Composio Connect setup guidance to configure its endpoint and app connection, then review which actions your ChatGPT workspace permits. Composio also supports developer-built agent workflows through session-specific MCP endpoints. See sessions via MCP.

Comparison: Which method should you choose?
Feature | Official Atlassian MCP | Open-source | Composio Connect |
|---|---|---|---|
Confluence support | Cloud | Cloud and Server/Data Center | Confluence Cloud integration |
Who operates the connection | Atlassian | Your team | Composio |
Authentication | OAuth 2.1, with optional API token | Depends on deployment and configuration | Managed app authorization |
Maintenance | Atlassian-hosted | Self-managed | Managed connection layer |
Governance | Atlassian permissions, scopes, and admin policies | Your infrastructure and client controls | Standard: Firewall and SSO. Enterprise adds audit logs, SIEM streaming, and SCIM provisioning |
Other applications | Atlassian products and supported services | Jira and Confluence | Multiple third-party applications |
Best fit | Teams using Confluence Cloud | Teams requiring self-hosting or Data Center | Teams connecting Confluence alongside other apps |
What your admin needs to approve
Before a team-wide connection goes live, ask your Atlassian administrator to review:
Client access. Open Atlassian Administration > Connectors > Atlassian MCP server and verify that your AI client's domain is allowed under the organization's OAuth domain policy. Atlassian-supported domains may be allowed by default, depending on policy.
Authentication. OAuth 2.1 is the default. API-token authentication can be enabled or disabled separately. It is relevant for service-style tools and does not use the OAuth domain allowlist.
Permission groups and scopes. Atlassian MCP v2 uses
read:confluence:agent-interface,write:confluence:agent-interface, andsearch:confluence:agent-interfacefor the corresponding Confluence read, write, and search groups. Grant only what the use case needs.Existing restrictions. Confirm Confluence space permissions and any applicable IP allowlists or organization security policies.
For the precise admin controls and current UI, follow Atlassian's policy documentation and supported tool permissions.
When the official server is enough, and when it isn’t
If your team only needs Claude or ChatGPT to read and edit Confluence Cloud, start with Atlassian’s official MCP server. It’s free, hosted by Atlassian, and governed by the admin controls you already use. The picture changes when Confluence is one of several tools in the workflow. An assistant that drafts a spec in Confluence, files the tickets in Jira, posts a summary in Slack and opens a branch in GitHub needs a separate connection for each non-Atlassian app, and each one brings its own authentication and upkeep.
Composio Connect exposes Confluence, Slack, GitHub, and other connected apps through one MCP URL and handles OAuth token refresh for each, so you don’t build or maintain separate auth flows per app. If you’re a developer building your own agents or customer-facing products, session-specific MCP endpoints let each end user connect their own accounts, without your team building and maintaining OAuth flows for every integration. In short: choose Atlassian’s server for Confluence alone, mcp-atlassian if you must self-host or run Data Center, and Composio when Confluence is part of a broader set of apps.
Security patterns for production use
Use approved authentication
Prefer OAuth 2.1 for interactive users, and enable API-token access only when your organization's policy and workflow call for it. Protect secrets and avoid placing keys in shared files.
Apply least privilege
Give read and search access to assistants that only need to answer questions. Grant write:confluence:agent-interface only for approved edit workflows, and retain human review for sensitive changes.
Respect access boundaries
MCP permissions work alongside Confluence permissions and organization-level access controls. Test with accounts representing ordinary users, not only administrators.
Review logs and policies
Atlassian's audit logs cover activity on the Confluence side. For Composio Connect, the Standard plan includes Firewall and single sign-on (SSO). The Enterprise plan adds audit logs, SIEM streaming, and SCIM provisioning, so you can send connection activity to your existing monitoring stack. Zero data retention (ZDR) is included on both Standard and Enterprise plans listed on the Composio Connect pricing page.
Troubleshooting authentication errors
Symptom | What to check | Next step |
|---|---|---|
Authentication fails or expires | OAuth consent, disconnected account, or revoked token | Reconnect using the supported authorization flow |
A page or space is inaccessible | The connected user's Confluence permissions | Verify access in the Confluence UI before retrying |
The client cannot connect | MCP URL, client support, domain policy, or IP allowlist | Confirm the |
Expected tools do not appear | MCP v2 tool discovery or the client's cached connection | Use tool discovery, refresh the connection, and inspect available tools |
For advanced troubleshooting, consult Atlassian's troubleshooting and verification guide. Developers can also use the MCP Inspector to inspect a compatible MCP server. Its package is @modelcontextprotocol/inspector. This is a general MCP diagnostic tool, not an Atlassian-specific doctor command.
Verify your connection works
Check connection status in your Claude or ChatGPT client’s connector settings.
Confirm tools or available actions. With Atlassian MCP v2, a client may show primary tools first and offer discovery for additional ones. Look for
getConfluenceContent,searchConfluence, or their discoverable equivalents.Test a read. Ask: “Summarize the Confluence page titled [Page name] that I have permission to access.”
Test a write only if authorized. Ask the assistant to create a clearly labeled test page in an approved space, confirm that the content is correct, then remove the test page according to your team’s process.
Connect Confluence to your AI workflow
For teams using Confluence Cloud, the official Atlassian MCP server is the most direct place to start. The open-source server is an option if you need to run your own connection or use Server/Data Center. Composio Connect is worth evaluating if your workflow spans Confluence and other business apps and you want a managed connection layer.
Book a call to walk through your security requirements before the next enterprise review.
FAQs
Does the Atlassian MCP server work with Confluence Data Center?
No. Atlassian's official hosted MCP server is designed for Atlassian Cloud. For Confluence Server or Data Center, evaluate the community-maintained mcp-atlassian project and confirm support for your version.
What OAuth scopes does the Confluence MCP connection require?
Atlassian MCP v2 documents read:confluence:agent-interface, write:confluence:agent-interface, and search:confluence:agent-interface for its Confluence permission groups. The groups needed depend on whether the workflow only reads and searches content or also creates and edits it.
How often do I need to re-authenticate the Confluence connection?
That depends on the authentication method, your identity provider, and your organization's security policies. Managed OAuth connections can refresh credentials, but you may need to reconnect if access is revoked, a connection becomes unhealthy, or organizational settings change. Check the actual token policy rather than assuming a universal expiration period.
Can ChatGPT edit Confluence pages directly?
On ChatGPT Plus or Pro, use a custom MCP connection to look up Jira issues or retrieve Confluence pages. Editing Atlassian content through custom MCP requires a Business, Enterprise, or Edu workspace. The connected server must offer the write action, the workspace must allow it, and the user must have the relevant Atlassian permissions. ChatGPT may ask for confirmation before making the change. Deep Research uses custom MCP connections for retrieval only, as explained in OpenAI’s guide to custom MCP connections.
What is the difference between Confluence MCP and the Confluence REST API?
The REST API is an application programming interface developers call directly. MCP exposes tools and workflows to AI clients. The two are not mutually exclusive: Atlassian MCP v2 already provides page operations, searches, attachments, blog posts, version history, and content restrictions, so these are not automatically reasons to abandon MCP. Use the REST API when you need a specific endpoint or implementation behavior that your MCP connection does not provide.
Does Composio store my Confluence data?
Composio provides managed authentication and connected-app access, but data retention depends on the specific product, plan, and configuration. Review Composio's current security and data-handling documentation and confirm contractual requirements before transmitting sensitive Confluence content. Do not assume that every Composio connection operates under a zero-data-retention arrangement.
Key terms glossary
Model Context Protocol (MCP): A shared protocol that lets AI assistants connect to outside tools and data.
Confluence Query Language (CQL): A query syntax for searching Confluence content using structured search expressions.
OAuth 2.1: The authorization approach used by Atlassian's hosted MCP server to grant scoped access to connected applications.
Permission group: A set of MCP tools sharing an access category, such as reading, writing, or searching Confluence content. Atlassian admins can control access at the group level.
Role-based access control (RBAC): A method for assigning application or action access based on user or team roles. The specific controls vary by platform.
