ar-init-mcps

name	ar-init-mcps
description	Initialize issue-tracker integration for Claude Code. For GitHub Issues (the default) this verifies the gh CLI is authenticated, no MCP needed. For Jira (Atlassian) or Linear it sets up the Model Context Protocol (MCP) server. Say "ar-init-mcps" or "setup mcps".
disable-model-invocation	true

Initialize Issue-Tracker Integration

Set up the integration for the project's configured issue tracker.

When to Use

• First time wiring up the project's issue tracker
• ar-taskflow needs ticket access but the integration isn't configured
• User says "ar-init-mcps", "setup mcps", or "configure mcp"

Step 0: Resolve the Tracker (do this FIRST)

Read tracker.type from .claude/config_hints.json (under project). Branch on it - this is the same tracker seam as the Tracker Dispatch Table in <standards_location>/mcp-integration.md (resolve standards_location from .claude/config_hints.json):

TRACKER=$(jq -r '.project.tracker.type // "github"' .claude/config_hints.json 2>/dev/null)

tracker.type	What this skill does
github (default)	No MCP needed, GitHub Issues uses the gh CLI. Just verify gh auth status (see below).
jira	Set up the Atlassian MCP server (Jira / Confluence), see the Atlassian section.
linear	Set up the Linear MCP server, then follow the same dispatch rows.
none	Nothing to configure, identifiers are managed manually.

GitHub Issues path (tracker.type: github)

No MCP server is required. Verify the gh CLI is installed and authenticated:

gh auth status

• If authenticated, you're done, ar-taskflow uses gh issue ... directly.
• If not, run gh auth login and re-check.
• If gh is missing, install it (brew install gh / your platform's package manager).

The rest of this skill is the MCP setup path, used only when tracker.type is jira (Atlassian) or linear.

What are MCP Servers?

MCP servers allow Claude Code to:

• Fetch Jira tickets and issues
• Read/write Confluence pages
• Access other external services
• Integrate with your development tools

Checking Current MCP Status

Before setting up, check if MCP servers are already configured:

claude mcp list

If the command returns servers, MCP is already configured. Ask user if they want to add more.

Available MCP Servers

1. Atlassian (Jira & Confluence)

What it provides:

• Read Jira tickets and issues
• Create/update Jira issues and descriptions
• Add comments to Jira tickets
• Read Confluence pages
• Search across Jira and Confluence

ar-taskflow Integration:

• Start tasks from Jira tickets (ticket-first approach)
• Automatically update ticket description when task completes
• Archive original description to comments
• Add completion status and branch info

Setup command:

claude mcp add --scope user --transport http atlassian https://mcp.atlassian.com/v1/mcp

After setup:

1. You'll be prompted to authenticate via browser
2. Grant permissions to Claude Code
3. Server will be available immediately

Enables in ar-taskflow:

• Ticket-first approach: Start tasks from Jira URLs
• Auto-fetch ticket descriptions to raw_prompt.md
• Auto-update tickets on completion with:
  • Original description archived to comments
  • Updated description from ticket.md
  • Completion status and branch info

2. Linear (tracker.type: linear)

Configure the Linear MCP server, then drive it through the same dispatch rows in <standards_location>/mcp-integration.md (fetch / search / create / update / link).

3. Future MCP Servers

As more MCP servers become available, add them here:

• Slack MCP (when available)

(GitHub Issues needs no MCP, it uses the gh CLI; see Step 0.)

Setup Flow

Step 1: Check Existing Configuration

claude mcp list

Step 2: Add Atlassian MCP

If not already configured:

To integrate with Jira and Confluence, run:

  claude mcp add --scope user --transport http atlassian https://mcp.atlassian.com/v1/mcp

This will:
1. Open your browser for authentication
2. Ask for Jira/Confluence permissions
3. Enable Claude to read/write tickets and pages

After setup, you can:
- Fetch Jira tickets by URL or ID
- Create tasks from Jira descriptions
- Search Confluence documentation
- Create/update Jira issues directly

Step 3: Verify Setup

After user runs the command:

claude mcp list

Should show:

atlassian (http://...)

Step 4: Test Connection

Try fetching a test resource to verify:

Ask user: "Do you have a Jira ticket URL I can test with?"

If yes, attempt to fetch it using MCP tools.

Troubleshooting

"Command not found: claude mcp"

User needs to update Claude Code CLI:

npm update -g @anthropic-ai/claude-code

"Authentication failed"

1. Check browser didn't block the auth popup
2. Try running the add command again
3. Ensure user has access to Atlassian instance

"Permission denied"

MCP scope might be wrong. Use --scope user for user-level config.

Integration with Task Flow

When ar-taskflow needs tracker access (for the ticket-first approach), it resolves tracker.type first:

1. github, verify gh auth status; no MCP step.
2. jira / linear, check the MCP server is configured; if not, invoke this skill to set it up, then continue.
3. none, skip; identifiers are managed manually.

Quick Reference

Command	Purpose
claude mcp list	Show configured servers
claude mcp add ...	Add new server
claude mcp remove <name>	Remove server
claude mcp test <name>	Test server connection

Notes

• MCP configuration is stored at user level (not project level)
• Once configured, available across all projects
• Credentials are managed securely by Claude Code
• Can add multiple MCP servers simultaneously