Quick Start

⚡ Quick Start Guide (5 Minutes)

Get up and running with mcp-server-decisions in 5 minutes.

---

📦 1. Installation

pip install -e .

---

⚙️ 2. Configuration

Add the stdio MCP server entry to your client configuration (e.g. Claude Desktop / OpenCode):

{
  "mcpServers": {
    "mcp-server-decisions": {
      "type": "stdio",
      "command": "mcp-server-decisions"
    }
  }
}

---

📝 3. Record Your First Decision

Ask your AI assistant or agent:

"Record this decision: we're using DuckDB for caching because it's fast and requires no external infra. Alternatives considered: Redis, SQLite, Elasticsearch. Predict p99 latency will drop below 200ms and storage cost will stay under $50/month."

The tool generates an immutable record:

• Problem: Query latency exceeds SLA
• Solution: DuckDB caching
• Rejected: Redis, SQLite, Elasticsearch
• Technologies: duckdb
• Predictions: LATENCY (p99 < 200ms), COST (< $50/month)

Returns: DEC-2026-0001 with prediction IDs (PRD-2026-0001, PRD-2026-0002).

---

🎯 4. Record the Outcome

After measuring in production or running benchmarks:

"We measured p99 latency at 180ms with DuckDB. Record this outcome with 95% accuracy."

Returns: SUCCESS ✅ (95% accuracy score mapped).

---

📊 5. Check Technology Performance

Run the built-in report script to inspect aggregated metrics:

python3 scripts/technology_performance_report.py

Output:

technology: duckdb | successful: 1 | failed: 0 | avg_accuracy: 95.0% | confidence: LOW

---

📌 Core Concepts At a Glance

Concept	Description
Decision ID (DEC-YYYY-NNNN)	Unique identifier for architectural choices
Prediction ID (PRD-YYYY-NNNN)	Measurable claim attached to a decision
Accuracy Score (0–100)	Measured proximity of actual outcome to initial prediction
Outcome Gate	In-band nudge in response payloads preventing unvalidated leaks

---

Made for AI agents. Built for teams. Learn from every decision.