MCP Atlassian

To deploy the MCP Atlassian server to Kubernetes using Atlassian Cloud API tokens, create a values.yaml file containing your Confluence and Jira credentials, then install the chart using Helm.

# Create values file
cat > my-values.yaml <<YAML
confluence:
  url: "https://your-company.atlassian.net/wiki"
  username: "your.email@company.com"
  apiToken: "your_confluence_api_token"

jira:
  url: "https://your-company.atlassian.net"
  username: "your.email@company.com"
  apiToken: "your_jira_api_token"
YAML

# Install the chart
helm install mcp-atlassian ./mcp-atlassian -f my-values.yaml

For multi-tenant applications where each user connects to their own Atlassian cloud instance, enable minimal OAuth mode and have users provide authentication via HTTP headers.

```bash
# 1. Enable minimal OAuth mode
# Using uvx
ATLASSIAN_OAUTH_ENABLE=true uvx mcp-atlassian --transport streamable-http --port 9000

# Or using Docker
docker run -e ATLASSIAN_OAUTH_ENABLE=true -p 9000:9000 \
  ghcr.io/sooperset/mcp-atlassian:latest \
  --transport streamable-http --port 9000

User Authentication Headers:

• Authorization: Bearer <user_oauth_token>
• X-Atlassian-Cloud-Id: <user_cloud_id>

To run the server using streamable-http transport, use the following commands and configure your IDE with the /mcp endpoint.

# Using uvx
uvx mcp-atlassian --transport streamable-http --port 9000 -vv

# Or using Docker
docker run --rm -p 9000:9000 \
  --env-file /path/to/your/.env \
  ghcr.io/sooperset/mcp-atlassian:latest \
  --transport streamable-http --port 9000 -vv

IDE Configuration:

{
  "mcpServers": {
    "mcp-atlassian-service": {
      "url": "http://localhost:9000/mcp"
    }
  }
}

The recommended way to run MCP Atlassian without a permanent installation is using uvx. This downloads the tool on first use and caches it for subsequent runs.

To run the help command directly:

uvx mcp-atlassian --help

To run with a specific Python version:

uvx --python=3.12 mcp-atlassian --help

If you don't want to write CQL, you can use confluence_search in two ways:

1. Simple Text Search: Pass plain text (e.g., project documentation architecture). confluence_search will automatically use siteSearch to mimic the Confluence web UI search.
2. Relevance-Ranked Search: Explicitly use siteSearch with the ~ operator for relevance-ranked results (e.g., siteSearch ~ "important concept").

JQL (Jira Query Language) is used by the jira_search tool to find issues. The basic syntax follows the pattern: field operator value.

project = "PROJ" AND status = "In Progress"

You can test your mcp-atlassian configuration interactively using the MCP Inspector.

Using uvx:

npx @modelcontextprotocol/inspector uvx mcp-atlassian

Using a local development version:

npx @modelcontextprotocol/inspector uv --directory /path/to/mcp-atlassian run mcp-atlassian

To use MCP Atlassian with an IDE, you must add it to the IDE's MCP configuration.

Claude Desktop Configuration Locations:

• Windows: %APPDATA%\Claude\claude_desktop_config.json
• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Linux: ~/.config/Claude/claude_desktop_config.json

Cursor: Navigate to Settings → MCP → + Add new global MCP server.

When writing JQL for jira_search:

• Deterministic Results: Always include ORDER BY in your queries to ensure predictable results.
• Relative Dates: Use relative date strings like "-7d", "-1w", or "-1M" instead of absolute dates to make queries reusable.
• Function Compatibility: Be aware that some functions (e.g., issueHistory()) may be Cloud-only. Verify support with your specific Jira version.

Organize Confluence content using a hierarchy:

1. Create a parent page: Use confluence_create_page with a space_key, title, and content.
2. Create child pages: Use confluence_create_page and provide the parent_id of the parent page.
3. Organize with labels: Use confluence_add_label to attach metadata to a page via its page_id.