MCP server providing Sprint Protocol tooling for LLM-driven development workflows
sprint-mcp is a Model Context Protocol (MCP) server that implements the Sprint Protocol, enabling structured, traceable development workflows for LLM-powered coding assistants like Claude.
Sprint Protocol is a structured methodology for LLM-driven development that provides:
- Phased execution (Plan → Approve → Implement → Validate → Verify → Publish → Retro → Learn)
- Git worktree isolation for each sprint
- Comprehensive artifacts (implementation plans, verification reports, retrospectives)
- Full traceability through request logs
- Quality gates ensuring code quality and testing
Model Context Protocol enables Claude Desktop to interact with external tools and services. sprint-mcp exposes Sprint Protocol tools as MCP endpoints.
- Sprint Lifecycle Management - Start, update, complete, and track sprints
- Git Worktree Integration - Isolated development environments per sprint
- Artifact Generation - Automated creation of plans, reports, retrospectives
- Sprint Index Management - Track all sprints across your projects
- Archive System - Organize completed sprints by year, free up active workspace
- Knowledge Extraction - Automatically extract lessons, patterns, and metrics from completed sprints
- Auto-Archive - Intelligent archival based on age, count, or hybrid criteria
- Cleanup Tools - Remove completed sprint worktrees safely
- Validation & Verification - Quality gates and artifact validation
npm install -g sprint-mcpPros: Simple configuration, faster startup, works offline
No installation needed - use directly with npx.
Pros: Always latest version, no global packages, easy to try
npm install --save-dev sprint-mcpPros: Project-specific version, locked in package.json
Add to your Claude Desktop configuration file:
Config Location:
- macOS:
~/Library/Application Support/Claude/claude_desktop_config.json - Windows:
%APPDATA%\Claude\claude_desktop_config.json - Linux:
~/.config/Claude/claude_desktop_config.json
{
"mcpServers": {
"sprint-mcp": {
"command": "sprint-mcp"
}
}
}{
"mcpServers": {
"sprint-mcp": {
"command": "npx",
"args": ["-y", "sprint-mcp"]
}
}
}{
"mcpServers": {
"sprint-mcp": {
"command": "sprint-mcp",
"env": {
"SPRINT_ROOT": "/path/to/your/project"
}
}
}
}After configuration:
- Restart Claude Desktop completely (Cmd+Q / File → Exit)
- Wait 5-10 seconds for MCP servers to initialize
- Verify tools appear in Claude
sprint-mcp provides the following MCP tools:
-
start-sprint- Initialize a new sprint with manifest and directory structure- Creates sprint directory in
planning/ - Sets up git worktree for isolated development
- Generates sprint manifest with metadata
- Creates sprint directory in
-
check-sprint-status- Verify current sprint state- Returns active sprint information
- Checks for conflicts (multiple active sprints)
- Validates sprint index integrity
-
update-sprint-status- Update sprint status and metadata- Status transitions: planning → in-progress → validating → verifying → published → complete
- Updates both manifest and sprint index
- Atomic operations for consistency
-
complete-sprint- Complete sprint with validation- Validates required artifacts (verification-report.md, retro.md, key-learnings.md)
- Supports normal and forced completion modes
- Updates status and timestamps
regenerate-sprint-index- Rebuild sprint index from manifests- Scans all sprint manifests in
planning/ - Rebuilds
planning/sprint-index.yaml - Validates index integrity
- Recovers from index corruption
- Scans all sprint manifests in
-
archive-sprint- Archive completed sprints to organized storage- Moves sprints from
planning/active/toplanning/archive/{year}/ - Updates sprint index with new location
- Triggers knowledge extraction (if enabled)
- Dry-run mode for preview
- Moves sprints from
-
auto-archive-sprints- Automatically archive eligible sprints- Age criteria: Archive sprints older than N days
- Count criteria: Keep only N most recent sprints
- Hybrid criteria: Must meet both age AND count thresholds
- Batch processing with error handling
- Configurable via
planning/archive-config.yaml
cleanup-sprint- Clean up completed sprint worktrees- Preview mode: Shows what will be deleted
- Execution mode: Removes git worktrees
- Safety checks: Only cleans completed sprints
- Force option: Override uncommitted changes check
In Claude Desktop:
Start a new sprint to implement user authentication
Claude will use the start-sprint tool to:
- Check no active sprints exist
- Generate sprint ID (e.g.,
sprint-12-abc123) - Create
planning/sprint-12-abc123/directory - Create git worktree in
.worktrees/sprint-12-abc123/ - Generate sprint manifest
- Update sprint index
Check the current sprint status
Returns information about active sprints and index health.
Complete the current sprint
Claude validates artifacts and marks sprint as complete.
Clean up completed sprint worktrees
Removes git worktrees for completed sprints, freeing disk space.
sprint-mcp implements the Sprint Protocol defined in AGENTS.md. Key concepts:
- Plan - Create implementation plan, get user approval
- Implement - Execute planned work, log all changes
- Validate - Run validation script, verify deliverables
- Verify - Create verification report, document gaps
- Publish - Create GitHub PR, publish deliverables
- Retro - Reflect on what went well / what didn't
- Learn - Extract transferable learnings
Every sprint creates:
sprint-manifest.yaml- Sprint metadata and statusimplementation-plan.md- Detailed execution planrequest-log.md- All prompts and changesvalidate_deliverable.sh- Executable validation scriptverification-report.md- Completed/partial/deferred itemsretro.md- What worked, what didn'tkey-learnings.md- Lessons for future sprints
Note: publication.yaml was deprecated in Protocol v2.5. PR URL and publication metadata are now tracked in sprint-manifest.yaml.
Each sprint uses an isolated git worktree:
- Main worktree stays on main branch
- Sprint worktree on feature branch
- Independent working directories
- Separate git operations
- Easy cleanup after completion
Completed sprints are organized in an archive hierarchy:
- Active sprints:
planning/active/{sprint-id}/ - Archived sprints:
planning/archive/{year}/{sprint-id}/ - Auto-archival: Configurable age/count/hybrid criteria
- Knowledge extraction: Automatic extraction of lessons and patterns
- Index tracking: All sprints remain in
planning/sprint-index.yaml
The system automatically builds a knowledge base from sprint artifacts:
- Lessons learned: Extracted from
key-learnings.md, retrospectives - Patterns: Successful approaches from "what went well" sections
- Anti-patterns: Things to avoid from "what to improve" sections
- Metrics: Sprint duration, effort, velocity tracking
- Deduplication: Similar knowledge merged with frequency tracking
- Storage:
planning/knowledge/knowledge-base.yaml
- Node.js: v18.0.0 or higher
- npm: v8.0.0 or higher
- Git: v2.20 or higher (for worktree support)
- Claude Desktop: Latest version
- Verify configuration file exists and is valid JSON
- Check Claude Desktop logs for errors
- Ensure sprint-mcp is installed (
which sprint-mcpornpx -y sprint-mcp) - Restart Claude Desktop completely
Global installation:
npm install -g sprint-mcp
which sprint-mcpnpx usage:
npx -y sprint-mcp # Test it worksFix npm permissions:
mkdir ~/.npm-global
npm config set prefix '~/.npm-global'
echo 'export PATH=~/.npm-global/bin:$PATH' >> ~/.bashrc
source ~/.bashrc
npm install -g sprint-mcpCheck Node.js version:
node --version # Must be v18+Run manually to see errors:
sprint-mcpnpm update -g sprint-mcpNo action needed - npx always uses latest version.
To pin version:
{
"args": ["-y", "sprint-mcp@0.1.0"]
}npm update sprint-mcp- Sprint Protocol: See
AGENTS.mdin package or GitHub - Installation Guide: Full installation guide
- Examples: GitHub examples directory
git clone https://github.com/cnavta/sprint-mcp.git
cd sprint-mcp
npm install
npm run build
npm link # Use locallynpm test # Run all tests
npm run test:coverage # With coverage reportCurrent test coverage: 310 tests passing across 18 test suites
Contributions welcome! Please:
- Fork the repository
- Create a feature branch
- Follow Sprint Protocol (see AGENTS.md)
- Include tests for new functionality
- Submit pull request
See CLAUDE.md for development guidelines.
MIT License - see LICENSE file for details
- npm Package: https://www.npmjs.com/package/sprint-mcp
- GitHub Repository: https://github.com/cnavta/sprint-mcp
- Issues: https://github.com/cnavta/sprint-mcp/issues
- MCP Specification: https://modelcontextprotocol.io/
- Built with Model Context Protocol SDK
- Designed for use with Claude Desktop
- Inspired by structured development methodologies
- Documentation: See installation guide
- Issues: GitHub Issues
- Discussions: GitHub Discussions