Skip to main content
MCP Atlassian supports four authentication methods depending on your Atlassian deployment type. The simplest method for Atlassian Cloud users.
2

Generate Token

Click Create API token, give it a name
3

Copy Token

Copy the token immediately - you won’t see it again
Environment variables:

Personal Access Token (Server/Data Center)

For Server or Data Center deployments.
1

Access Profile

Go to your profile (avatar) → Profile → Personal Access Tokens
2

Create Token

Click Create token, name it, set expiry
3

Copy Token

Copy the token immediately
Environment variables:
For multi-user HTTP deployments where clients send their own PATs, follow the Server/DC per-request authentication setup. That setup uses non-empty server-side PAT values only to initialize the service configuration; each request’s Authorization: Token ... value replaces them.
Enterprise CA certificates in the OS trust store (Windows Certificate Store, macOS Keychain, Linux system CAs) are trusted automatically. For self-signed certificates not in the OS trust store, set JIRA_SSL_VERIFY=false and/or CONFLUENCE_SSL_VERIFY=false.

Client Certificate (mTLS) (Server/Data Center)

Use mutual TLS when your Jira or Confluence Server/Data Center instance requires a client certificate. This authentication method does not require a username, API token, or personal access token. Set *_CLIENT_CERT to either a combined certificate-and-private-key PEM, or set it with the matching *_CLIENT_KEY when the certificate and key are separate files:
In the example, Jira uses separate certificate and key files, while Confluence uses a combined PEM. For a combined PEM, omit *_CLIENT_KEY. The matching JIRA_* and CONFLUENCE_* variables configure each service independently.
Encrypted private keys are not supported. If your key is passphrase-protected, decrypt it first:
JIRA_CLIENT_KEY_PASSWORD and CONFLUENCE_CLIENT_KEY_PASSWORD are recognised environment variables but will raise a configuration error at startup if set to a non-empty value.
If your Server/DC instance uses a self-signed or internal CA certificate alongside mTLS, set JIRA_SSL_VERIFY=false and/or CONFLUENCE_SSL_VERIFY=false as well.
When using mTLS as the sole authentication method, MCP clients that connect without an Authorization header (such as a locally running OpenCode instance) will be rejected by default. Set ALLOW_GLOBAL_CRED_FALLBACK=true to allow the server to use the configured certificate credentials for requests that carry no per-request auth header:
Only set this when the server is not exposed to untrusted callers, as it allows any unauthenticated client to transact under the operator’s certificate identity.

OAuth 2.0 (Cloud) - Advanced

OAuth 2.0 provides enhanced security features but requires more setup. For most users, API Token authentication is simpler and sufficient.

Setup Steps

1

Create OAuth App

Go to Atlassian Developer Console and create an “OAuth 2.0 (3LO) integration” app
2

Configure Permissions

Add scopes for Jira/Confluence as needed
3

Set Callback URL

Set to http://localhost:8080/callback
4

Run Setup Wizard

5

Complete Authorization

Follow prompts for Client ID, Secret, URI, and Scope, then complete browser authorization
Environment variables (after setup):
Include offline_access in your scope to allow automatic token refresh.

Confluence Cloud granular scopes

On Cloud, OAuth requests go through the api.atlassian.com gateway. Atlassian has removed most Confluence v1 REST endpoints there (410 Gone), so the server uses the v2 REST API for those operations. v2 endpoints accept only granular scopes: with classic scopes such as read:confluence-content.all alone they answer 401 scope does not match. Add the granular scopes to your app in the developer console (Permissions, Confluence API, Granular scopes), list them in ATLASSIAN_OAUTH_SCOPE, and authorize again so the token carries them. Keep the classic scopes as well. Search and a few writes (adding a label, uploading an attachment) still use v1 endpoints that remain on the gateway. A read-only setup needs only the read: rows.

MCP OAuth Proxy (DCR + Discovery)

Enable this when running a remote MCP endpoint that should onboard MCP clients through the standard OAuth discovery/DCR flow (401 challenge, /.well-known/*, /register, /authorize, /token, callback).
This mode is opt-in. Existing API token, PAT, and header-based OAuth flows continue to work without enabling the proxy.
The callback configured in ATLASSIAN_OAUTH_REDIRECT_URI is for the upstream Atlassian identity provider, not a dynamically registered MCP client. A DCR client’s redirect_uri must be different from this OAuth proxy IdP callback. Registration is rejected with invalid_redirect_uri if the two callbacks collide, preventing a callback loop.

Bring Your Own Token (BYOT)

If you manage OAuth tokens externally (e.g., through a central identity provider):
Token refresh is your responsibility - the server does not handle it for BYOT.

Multi-Cloud OAuth

For multi-tenant applications where users provide their own OAuth tokens:
  1. Enable minimal OAuth mode:
  2. Users provide authentication via HTTP headers:
    • Authorization: Bearer <user_oauth_token>
    • X-Atlassian-Cloud-Id: <user_cloud_id>
See HTTP Transport for more details on multi-user authentication.