atlassian-mcp-server

Atlassian Rovo MCP Server skills

The skills in this directory target the Atlassian Rovo MCP v2 endpoint, https://mcp.atlassian.com/v2/mcp. They are what every packaged plugin installs, and each one lives at skills/<name>/SKILL.md per the Agent Skills convention.

The previous generation is archived in v1/, unchanged, for anyone still pointed at a v1 endpoint (/v1/mcp/authv2 or /v1/mcp). No plugin installs the archived skills — plugin discovery only looks one level deep for a SKILL.md, so v1/ is skipped. Set them up manually if you need them.

What changed between v1 and v2

A skill written for one generation will fail against the other. The most consequential differences:

• Not every tool is directly callable. v2 exposes a small set of primary tools in the client’s tool list; the rest are reached through the discover and execute meta-tools. Each skill that needs one explains the convention in its Calling non-primary tools section.
• Confluence tools were renamed from *Page to *Content — getConfluencePage → getConfluenceContent, createConfluencePage → createConfluenceContent, updateConfluencePage → updateConfluenceContent, and searchConfluenceUsingCql → searchConfluence. The v1 names survive only as search-recall synonyms, not callable operations.
• Confluence read and write contracts changed. Reads default to a summary and need detail="full" for the body; document edits require a snapshotToken from the preceding read; writes must honor space instructions via getConfluenceSpace; and create/update take a parent object plus a required contentType, with body as {format, value}.
• Some Jira tools were renamed — getVisibleJiraProjects → listJiraProjects, getJiraProjectIssueTypesMetadata → listJiraProjectIssueTypesMetadata, addCommentToJiraIssue → addOrEditJiraIssueComment.
• Jira create parameters differ — createJiraIssue takes issueType and assignee, not issueTypeName and assignee_account_id.

For the current tool reference, see Supported tools.

This site is open source. Improve this page.