-
Notifications
You must be signed in to change notification settings - Fork 0
Home
Your Name edited this page Jul 23, 2026
·
3 revisions
Architectural decision tracking with prediction validation and outcome gates.
| Topic | Description | Link |
|---|---|---|
| ⚡ Quick Start | 5-minute setup guide to record your first decision | Quick Start Guide |
| 🚀 Getting Started | Overview of capabilities & architecture | Getting Started |
| 📊 Outcome Gate Pattern | Why in-band nudges boost loop closure to 14.5% | Outcome Gate Pattern |
| ❓ FAQ | Common questions on storage, predictions & accuracy | FAQ |
mcp-server-decisions is a Model Context Protocol (MCP) server that empowers AI agents and engineering teams to:
- 📝 Record technical choices (problem, chosen solution, rejected alternatives, technologies).
- 🔮 Predict testable outcomes (latency thresholds, cost boundaries, reliability criteria).
- 📏 Measure production metrics after deployment.
- ✅ Validate predictions against empirical data (0–100 accuracy score).
- 🧠 Learn from accuracy trends to improve subsequent architectural recommendations.
All persisted in an append-only JSONL log. Zero database setup, zero schema migrations, stdlib Python only.
Decide ➔ Predict ➔ Implement ➔ Measure ➔ Validate ➔ Learn ➔ Next Decision
Traditional architectural decision records (ADRs) leak feedback:
- A decision is made and documented.
- System is built and deployed.
- Nobody checks if the original predictions held true.
- Future decisions reuse the tech without knowing past accuracy.
mcp-server-decisions solves this using Outcome Gates: in-band nudges in tool responses reminding agents and users of open predictions.
Query aggregated historical performance for any technology before making a decision:
technology: duckdb | successful: 12 | failed: 1 | avg_accuracy: 91.2% | confidence: HIGH
technology: redis | successful: 8 | failed: 3 | avg_accuracy: 78.3% | confidence: MEDIUM
Made for AI agents. Built for teams. Learn from every decision.