| name | atlassian-mcp |
| description | Interact with Jira tickets and Confluence wiki pages via the Atlassian MCP server. USE FOR: reading/creating/updating Jira issues, searching Jira with JQL, reading/creating/updating Confluence pages, adding comments. ALWAYS use when user references a Jira ticket (e.g. DSC-1986, B2C-12345) or Confluence page. |
Atlassian MCP
Prefer Atlassian MCP tools over opening Atlassian URLs in a browser or asking the user to paste content.
Tool Drift Guardrails
Use the Atlassian tool names exposed in the current session instead of assuming one exact MCP function surface. Names below are intent labels from common Atlassian MCP surfaces; if the host exposes namespaced or renamed tools, map the same Jira fetch, Confluence fetch, search, JQL/CQL, transition, and account lookup intents to the current tools.
Default Route
- If the user gives an exact Jira key (for example
DSC-1986), fetch it directly with the Jira issue fetch tool using issueIdOrKey when that parameter is exposed.
- If the user gives a Confluence URL, extract the site hostname as
cloudId and fetch the page directly if the URL contains a usable page ID.
- If the user needs discovery and you do not know whether the answer is in Jira or Confluence, start with the generic Atlassian search tool, then fetch the returned resource.
- Use JQL or CQL search tools only when you need precise filtering that generic search cannot express.
Jira Issue Reading
- For an exact Jira key like
DSC-2209, use getJiraIssue directly with issueIdOrKey: "DSC-2209".
- If
search returns a Jira issue ARI, use fetch with that ARI as the fallback route.
- Do not use Confluence page tools,
getIssueLinkTypes, or remote-link tools to read a Jira issue body; those are for pages or link metadata.
Non-Obvious Rules
- Prefer Markdown output when you need readable text. Use ADF only when rich formatting matters.
- When the user provides an Atlassian URL such as
https://site.atlassian.net/..., try site.atlassian.net as cloudId before calling getAccessibleAtlassianResources.
- Keep JQL/CQL searches small by default with
maxResults: 10 or limit: 10.
- Before transitioning a Jira issue, call the transition-list tool to get a valid transition ID.
- Before assigning a Jira issue by display name, call the account lookup tool.
- Confluence tiny-link IDs from
/wiki/x/<id> can be passed directly as pageId.
Guardrails
- Do not open Jira or Confluence links in a browser when MCP can fetch them directly.
- Do not start with broad search when the user already gave an exact key, page ID, or URL.
- Use
search as the default discovery tool; use JQL/CQL only for explicit or precision-filtered queries.
Viewing Jira Attachments (Images)
The Atlassian MCP tools cannot download attachment files directly. To view images:
- Fetch the ticket with the Jira issue fetch tool — the
attachment field has the download URL in content
- Download via
curl with basic auth using $JIRA_USER_EMAIL and $JIRA_API_TOKEN from ~/.zshrc
- Use
view_image to inspect the downloaded file
Procedure
- Use the Atlassian MCP tools exposed in the current session. If the host groups tools behind activation, activate the Jira, Confluence, or search category that matches the task before calling the exact operation.
- If needed, use
getAccessibleAtlassianResources to get the cloud ID.
- Call the appropriate tool for the operation.
- When the user references a Jira ticket ID (e.g., DSC-1986), fetch it directly - never try to open the URL in a browser.
- If the ticket has attachments with images, download and view them using the method above.