Devpipe MCP Server

# devpipe-mcp Roadmap ## v0.2.2 (Current) ✅ - 28 resources (complete documentation + intelligence) - 18 tools (full devpipe lifecycle) - 6 prompts (common workflows) - Intelligence features: flakiness, performance regressions, change correlation - **NEW:** Pipeline health scoring ## v0.3.0 - Advanced Analytics ### High Priority #### 1. `compare_runs` Tool ✅ DONE (v0.2.2) **Purpose:** Compare two pipeline runs to identify changes **Use case:** ``` compare_runs --run1 latest --run2 previous Output: - New failures: go-lint, security-scan - Performance regressions: unit-tests (+50%) - Fixed tasks: integration-tests - Performance improvements: build (-20%) ``` **Implementation:** ✅ Complete - Diff two run.json files - Compare task results, durations, metrics - Identify regressions and improvements - Support "latest" and "previous" shortcuts **Effort:** Low (2-3 hours) - COMPLETED #### 2. `predict_impact` Tool ✅ DONE (v0.2.2) **Purpose:** Predict which tasks will fail based on changed files **Use case:** ``` predict_impact Output: - Critical risk: integration-tests (score: 85) - 3 changed files match watchPaths - High correlation with past failures - 40% recent failure rate - Recommendation: Run high-risk tasks first: integration-tests,security-scan - Suggested command: devpipe --only integration-tests,security-scan ``` **Implementation:** ✅ Complete - Multi-factor risk scoring (watchPaths, correlation, failure rate) - Risk levels: critical (70+), high (50+), medium (30+), low - Actionable recommendations with suggested commands - Uses existing intelligence data **Effort:** Medium (4-5 hours) - COMPLETED ### Medium Priority #### 3. Enhanced `diagnose_failure` ⭐⭐⭐ **Purpose:** Deep failure analysis with pattern matching **Note:** LLMs already do basic diagnosis. This would add: - Known error pattern database - Historical fix suggestions - Automated log parsing **Use case:** ``` diagnose_failure --task go-build Output: - Error type: missing_import - Affected file: main.go:42 - Similar past failures: 2 (both fixed by adding imports) - Suggested fix: Add import "github.com/myproject/database" ``` **Implementation:** - Build error pattern library - Match current errors to patterns - Query historical fixes **Effort:** High (8-10 hours) #### 4. `prioritize_tasks` Tool ⭐⭐⭐ **Purpose:** Optimize task execution order for fast feedback **Use case:** ``` prioritize_tasks Output: 1. go-fmt (30s, 95% failure detection) 2. go-vet (45s, 90% failure detection) 3. unit-tests (2m, 85% failure detection) ... 10. e2e-tests (10m, 60% failure detection) Recommendation: Run tasks 1-5 first (3.5m, 92% coverage) ``` **Implementation:** - Risk scoring algorithm - Time vs value optimization - Dependency-aware ordering **Effort:** High (6-8 hours) ### Low Priority #### 5. `get_intelligent_context` ⭐⭐ **Purpose:** Synthesize all data for debugging **Note:** LLMs already do this naturally. Only add if we want pre-computed summaries. **Effort:** Medium (3-4 hours) ## v0.4.0 - Future Enhancements ### devpipe v0.2.0 Compatibility ✅ DONE (v0.2.2) **Completed changes:** - ✅ `metricsFormat` → `outputType` - ✅ `metricsPath` → `outputPath` - ✅ `artifacts/` → `outputs/` (run folder structure) - ✅ `repoRoot` → `projectRoot` **MCP updates completed:** 1. ✅ Updated `configure-metrics` prompt 2. ✅ Updated type definitions (DevpipeTask) 3. ✅ Updated all documentation 4. ✅ Minimum version now v0.2.0 **Effort:** Low (1-2 hours) - COMPLETED ## Future Considerations ### Machine Learning Features - Failure prediction models - Optimal test selection - Anomaly detection **Blocker:** Need more data and ML infrastructure ### Real-time Monitoring - WebSocket support for live updates - Streaming task output - Progress notifications **Blocker:** MCP protocol limitations ### Team Analytics - Multi-user failure patterns - Team velocity metrics - Collaboration insights **Blocker:** Requires team/org context ## Implementation Strategy ### Phase 1: Quick Wins (v0.3.0) 1. ✅ `get_pipeline_health` (DONE in v0.2.2) 2. `compare_runs` (2-3 hours) 3. `predict_impact` (4-5 hours) **Total:** 1 week ### Phase 2: Advanced Features (v0.3.1) 4. Enhanced `diagnose_failure` (8-10 hours) 5. `prioritize_tasks` (6-8 hours) **Total:** 2 weeks ### Phase 3: Maintenance (v0.4.0) - Devpipe field rename support - Documentation updates - Bug fixes **Total:** Ongoing ## Decision Criteria **Implement if:** - ✅ Uses existing data (no new data sources) - ✅ Provides value LLMs can't replicate - ✅ Solves common pain points - ✅ Low maintenance burden **Defer if:** - ❌ Duplicates LLM capabilities - ❌ Requires ML infrastructure - ❌ Niche use case - ❌ High complexity ## Community Input Want a feature prioritized? Open an issue with: - Use case description - Expected output format - Why LLMs can't do it already