Team Memory MCP — AI skill for Claude Code

Team Memory MCP

Shared team memory for AI coding agents. Bayesian confidence scoring with temporal decay. Works with **Claude Code**, **Devin**, **Cursor**, and any MCP-compatible client.

The Problem

AI coding agents learn valuable things during sessions:

• "Spring Boot: use @Transactional(readOnly = true) on read-only queries to avoid unnecessary write locks"
• "PostgreSQL: always add CONCURRENTLY to CREATE INDEX on tables over 100k rows to avoid locking production"
• "REST APIs: return 409 Conflict (not 400) when a resource already exists — the API gateway retries on 400"
• "Flyway: migration files must follow V{ticket}_{seq}__{description}.sql or CI rejects the PR"

But when the session ends, that knowledge disappears. The next session starts from zero.

Multiply that across a team of 10 engineers, each using AI agents independently, and you get the same mistakes rediscovered week after week. The same wrong `@Transactional` scope. The same locking `CREATE INDEX`. The same 400 vs 409 confusion.

The Solution

Team Memory gives AI agents a persistent, shared knowledge store where engineering patterns are stored, validated by the team, and ranked by confidence over time.

1. Engineers or AI agents store patterns they discover during development
2. Each pattern starts with a confidence of ~0.667
3. Confirmations from team members increase confidence; corrections decrease it
4. Patterns not confirmed for 90 days gradually lose confidence (temporal decay)
5. Search results are ranked by confidence — well-validated patterns surface first

What This Looks Like in Practice

"Spring Boot: use @Transactional(propagation = REQUIRES_NEW) for audit logging
 to ensure logs persist even if the parent transaction rolls back"
→ Confirmed 23 times | Confidence: 0.92

"PostgreSQL: always add CONCURRENTLY to CREATE INDEX on tables with 100k+ rows"
→ Confirmed 15 times | Confidence: 0.88

"REST APIs: return 409 Conflict (not 400) when resource a