Atlassian CLI MCP Server

MCP server that wraps the Atlassian CLI (acli) into two generic tools — same pattern as gh-cli-mcp-server.

Why?

The official Atlassian MCP server exposes 30+ individual tools. This server wraps the entire acli binary into just 2:

• call_acli — run any acli command
• suggest_acli_commands — get help with syntax

No need for a dedicated tool per operation. acli already covers Jira (work items, projects, boards, sprints) and Confluence (pages, spaces, blogs). This just gives MCP clients a way to execute those commands directly.

Why not just use acli through a shell tool?

You can — it works. This server just cleans things up:

• Parses acli output into structured JSON when possible, instead of raw terminal text
• Handles timeouts so commands don't hang forever
• Includes a suggest_acli_commands tool for when you're unsure about syntax
• Shows up as a dedicated "atlassian" capability instead of generic shell access
• Blocks shell injection (no ;, |, &&, etc.)

Same acli underneath, just a safer wrapper for MCP clients to work with.

Tools

• call_acli — Execute any acli command and get structured output
• suggest_acli_commands — Get command suggestions from a natural language query

Prerequisites

• Atlassian CLI (acli) installed and authenticated
• uv (recommended, for running with uvx)

Installation

pip install acli-mcp-server

Usage with MCP clients

{
  "mcpServers": {
    "atlassian": {
      "command": "uvx",
      "args": ["acli-mcp-server"]
    }
  }
}

Examples

Just ask your MCP client naturally — it'll use call_acli under the hood:

If the client isn't sure which command to use, it can call suggest_acli_commands with your prompt to get suggestions first.