This does not try to wrap Atlassian in a friendly abstraction — it hands the assistant CQL and JQL and lets it write the query, which is the right call when the person asking already thinks in those terms. What it does not do is write: no issue creation, no transitions, no page edits, so it is safe to point at a production instance and useful mainly for finding and reading.
A small MCP server that puts Atlassian's two query languages in front of an assistant: CQL against Confluence content and JQL against Jira issues. It also registers resource templates so a page or an issue can be addressed directly by space and title, or by project and issue key.
- Confluence content found by CQL query — pages, attachments and comments, filterable by space, with a result limit between 1 and 50 — `confluence_search`
- Jira issues found by JQL query, with the field list you want returned (summary, status, assignee, timestamps) and the same 1 to 50 limit — `jira_search`
- Direct addressing of a single item through resource templates: `confluence://{space_key}/pages/{title}` and `jira://{project_key}/issues/{issue_key}`
An Atlassian account and an API token for each product you want reached. Confluence takes `CONFLUENCE_URL`, `CONFLUENCE_USERNAME` and `CONFLUENCE_API_TOKEN`; Jira takes `JIRA_URL`, `JIRA_USERNAME` and `JIRA_API_TOKEN`. The npm package name is `mcp-confluence` (1.0.0); install with `npx -y @smithery/cli install @petrsovadina/mcp-atlassian --client claude`, or clone, run `npm install` and `npm run build`, then point your client at `build/index.js`. MIT licensed.
One command plus a key — npx -y @smithery/cli install @petrsovadina/mcp-atlassian --client claude, then supply credentials