MCP GitHub Issue Priority Server

A Model Context Protocol (MCP) server that enables AI assistants to manage GitHub issues with deterministic priority scoring and concurrency-safe selection.

Quick Start

# 1. Install from npm (recommended)
npm install -g mcp-git-issue-priority

# 2. Authenticate (if not already using GitHub CLI)
gh auth login

# 3. Add to Claude Code (~/.claude.json)

{
  "mcpServers": {
    "github-issue-priority": {
      "command": "mcp-git-issue-priority"
    }
  }
}

# 4. Restart Claude Code and verify
# The MCP tools should appear when you run /mcp

Features

• Priority-Based Issue Selection: Deterministic scoring algorithm ensures consistent issue prioritization across sessions
• Concurrency-Safe Locking: File-based atomic locking prevents multiple AI sessions from selecting the same issue
• Guided Workflow: 8-phase workflow (selection → research → branch → implementation → testing → commit → pr → review) with transition validation
• Automatic Labeling: Creates and manages priority (P0-P3), type (bug, feature, chore, docs), and status labels
• Stale Lock Detection: Automatically detects and cleans up locks from dead processes
• Audit Logging: JSON Lines logging for all operations with 30-day retention

Installation

Prerequisites

• Node.js 20+ - Download
• GitHub CLI (recommended) - Install and run gh auth login
  • Or: a GitHub personal access token with repo scope

Install from npm (recommended)

npm install -g mcp-git-issue-priority

Install specific version

npm install -g [email protected]

Install from source

git clone https://github.com/steiner385/mcp-git-issue-priority.git
cd mcp-git-issue-priority
npm install && npm link

Troubleshooting

| Issue | Solution | |-------|----------| | command not found: mcp-git-issue-priority | Ensure npm global bin is in your PATH: npm bin -g | | GitHub authentication required | Run gh auth login or set GITHUB_TOKEN | | Build errors during install | Ensure Node.js 20+ is installed: node --version |

Configuration

GitHub Authentication

The server supports two authentication methods:

Recommended: GitHub CLI (automatic)

If you have GitHub CLI installed and authenticated, the server automatically retrieves your token:

# One-time setup
gh auth login

This is the recommended approach - no manual token management required.

Alternative: Environment Variable

Set GITHUB_TOKEN with a personal access token that has repo scope:

export GITHUB_TOKEN="ghp_your_personal_access_token"

Or configure it in your MCP settings (see below).

Claude Code Configuration

Add to ~/.claude.json (global) or .claude/settings.json (project):

{
  "mcpServers": {
    "github-issue-priority": {
      "command": "mcp-git-issue-priority"
    }
  }
}

Using a personal access token instead of GitHub CLI? Add the token to the config:

{
  "mcpServers": {
    "github-issue-priority": {
      "command": "mcp-git-issue-priority",
      "env": {
        "GITHUB_TOKEN": "ghp_your_personal_access_token"
      }
    }
  }
}

Verify Installation

After restarting Claude Code:

1. Run /mcp to see available MCP servers
2. The github-issue-priority server should be listed with 13 tools
3. Try list_backlog on any repository to confirm it's working

Available Tools

create_issue

Create a new GitHub issue with mandatory priority and type labels.

Arguments:
  - title (required): Issue title
  - body (optional): Issue description
  - priority (required): P0 (critical) | P1 (high) | P2 (medium) | P3 (low)
  - type (required): bug | feature | chore | docs
  - repository (required): "owner/repo" format

select_next_issue

Select and lock the highest-priority issue from the backlog. Uses deterministic scoring to ensure consistent selection.

Arguments:
  - repository (required): "owner/repo" format
  - type (optional): Filter by issue type
  - maxPriority (optional): Only consider issues at or above this priority

list_backlog

List all open issues in priority order without acquiring locks (read-only).

Arguments:
  - repository (required): "owner/repo" format
  - type (optional): Filter by issue type
  - limit (optional): Maximum issues to return (default: 20)

advance_workflow

Advance the workflow to the next phase for a locked issue.

Arguments:
  - issueNumber (required): Issue number to advance
  - targetPhase (required): research | branch | implementation | testing | commit | pr | review
  - repository (required): "owner/repo" format
  - testsPassed (optional): Required when advancing to 'commit' phase
  - prTitle (optional): Required for 'pr' phase
  - prBody (optional): Required for 'pr' phase
  - skipJustification (optional): Required if skipping phases

get_workflow_status

Get the current workflow status for locked issues.

Arguments:
  - issueNumber (optional): Specific issue number
  - repository (optional): "owner/repo" format

release_lock

Release lock on an issue (on completion, abandonment, or merge).

Arguments:
  - issueNumber (required): Issue number
  - reason (required): completed | abandoned | merged
  - repository (required): "owner/repo" format

force_claim

Force claim an issue locked by another session (requires confirmation).

Arguments:
  - issueNumber (required): Issue number to claim
  - confirmation (required): Must be exactly "I understand this may cause conflicts"
  - repository (required): "owner/repo" format

sync_backlog_labels

Detect and optionally fix issues missing required priority/type/status labels.

Arguments:
  - repository (required): "owner/repo" format
  - mode (optional): "report" (default) to list issues, "update" to apply labels
  - defaultPriority (optional): P0 | P1 | P2 | P3 (defaults to P2)
  - defaultType (optional): bug | feature | chore | docs (defaults to feature)

In report mode, returns a list of all issues missing labels with details about what's missing.

In update mode, applies default labels to issues:

• Missing priority → priority:P2 (or specified default)
• Missing type → type:feature (or specified default)
• Missing status → status:backlog

get_pr_status

Check CI status, approval state, and merge state of a pull request.

Arguments:
  - repository (required): "owner/repo" format
  - prNumber (required): Pull request number to check

bulk_update_issues

Add/remove labels and close/reopen multiple issues at once.

Arguments:
  - repository (required): "owner/repo" format
  - issues (required): Array of issue numbers (1-50)
  - addLabels (optional): Labels to add
  - removeLabels (optional): Labels to remove
  - state (optional): "open" | "closed"

implement_batch

Start implementing a batch of N issues in priority order. Returns the first issue to implement.

Arguments:
  - repository (required): "owner/repo" format
  - count (required): Number of issues to implement (1-10)
  - includeTypes (optional): Only include these issue types
  - excludeTypes (optional): Exclude these issue types
  - maxPriority (optional): Only P0, P1, etc.

batch_continue

Continue batch implementation. Polls for PR merge, then returns next issue or completion.

Arguments:
  - batchId (required): Batch ID from implement_batch
  - prNumber (optional): PR number for current issue

get_workflow_analytics

Get time-based workflow analytics: cycle time, phase breakdown, aging reports.

Arguments:
  - repository (required): "owner/repo" format
  - period (optional): "7d" | "30d" | "90d" | "all" (default: 30d)

Priority Scoring Algorithm

Issues are scored using a deterministic formula:

score = (basePoints + ageBonus) * blockingMultiplier * blockedPenalty

• Base Points: P0=1000, P1=100, P2=10, P3=1
• Age Bonus: +1 point per day since creation (max 365)
• Blocking Multiplier: 1.5x for issues with "blocking" label
• Blocked Penalty: 0.1x for issues blocked by open parent issues (via GitHub sub-issues)
• Tiebreaker: Earlier creation date wins (FIFO)

Dependency Detection

Issues with open parent issues (using GitHub's sub-issues feature) are automatically deprioritized with a 0.1x penalty. This ensures that blocked work sinks to the bottom of the backlog until its dependencies are resolved. Once a parent issue is closed, the child issue's priority returns to normal.

Workflow Phases

1. selection: Issue selected and locked
2. research: Understanding the problem
3. branch: Feature branch created
4. implementation: Code changes in progress
5. testing: Running tests and validation
6. commit: Changes committed
7. pr: Pull request created
8. review: Awaiting review/merge

Data Storage

All data is stored locally in ~/.mcp-git-issue-priority/:

~/.mcp-git-issue-priority/
├── locks/          # Active lock files (.lockdata)
├── workflow/       # Workflow state files (.json)
└── logs/           # Audit logs (JSON Lines format)

Development

Build

npm run build

Test

npm test

Lint

npm run lint

Type Check

npm run typecheck

Releasing (Maintainers)

One-time Setup

1. Create an npm account at https://www.npmjs.com/
2. Go to https://www.npmjs.com/settings/tokens
3. Create an "Automation" token with "Publish" permission
4. Add to GitHub: Settings → Secrets → Actions → New secret named NPM_TOKEN

Creating a Release

# Patch release (bug fixes): 1.0.0 → 1.0.1
npm version patch -m "Release v%s"
git push && git push --tags

# Minor release (new features): 1.0.0 → 1.1.0
npm version minor -m "Release v%s"
git push && git push --tags

# Major release (breaking changes): 1.0.0 → 2.0.0
npm version major -m "Release v%s"
git push && git push --tags

Pushing a tag triggers the release workflow which:

• Runs lint, build, and tests
• Creates a GitHub release with the tarball attached
• Publishes to npm registry

Download Tracking

• GitHub: Releases page shows download count per .tgz asset
• npm: https://www.npmjs.com/package/mcp-git-issue-priority shows weekly downloads

License

MIT License - see LICENSE for details.

Contributing

Contributions are welcome! Please see CONTRIBUTING.md for guidelines.

Support

• Open an issue for bug reports or feature requests
• Check existing issues before creating new ones