---

CLAUDE.md and MemClaw work together to give Claude Code complete project context. This guide explains how to use both effectively — static context in CLAUDE.md, dynamic context in MemClaw.

CLAUDE.md + MemClaw: The Complete Context Stack for Claude Code

Claude Code has two built-in mechanisms for project context: CLAUDE.md and MCP tools. Used together, they give Claude everything it needs to be genuinely useful on your project.

Complete your context stack with MemClaw → memclaw.me

---

The Two Layers of Context

Layer 1: CLAUDE.md (Static Context)

CLAUDE.md is a Markdown file in your project root. Claude reads it automatically at every session start. It's your project brief — the stable context that doesn't change often.

What goes in CLAUDE.md:

• Tech stack and versions
• Architecture rules and conventions
• What NOT to do (rejected approaches)
• Key file locations

Characteristics:

• Static — you update it manually
• Always loaded — every session, every time
• Version controlled — part of your repo
• Shared automatically — everyone who clones the repo gets it

Layer 2: MemClaw (Dynamic Context)

MemClaw is a persistent workspace that grows over time. It stores context that changes — session history, evolving decisions, sprint goals.

What goes in MemClaw:

• Session history (what was built, when)
• Decision log (what was decided and why)
• Current sprint goals (updated each sprint)
• Known issues (updated as they're resolved)

Characteristics:

• Dynamic — grows automatically via /end command
• Loaded on demand — Claude queries it at session start
• Not version controlled — stored in MemClaw's cloud
• Shareable — team members can share a workspace

---

How They Work Together

Session Start
    ↓
Claude reads CLAUDE.md
    ↓ (stable context: tech stack, architecture rules)
Claude queries MemClaw workspace
    ↓ (dynamic context: session history, decisions, sprint goals)
Claude has complete project context
    ↓
Productive session
    ↓
/end command
    ↓
Claude writes session summary to MemClaw
    ↓
Context grows for next session

---

Setting Up the Complete Stack

CLAUDE.md

# Project Name

## Tech Stack
- Next.js 14 App Router, TypeScript, PostgreSQL 16
- Deployed on Railway; staging: staging.example.com

## Architecture Rules
- All DB access via Repository Pattern (/lib/repositories/)
- JWT auth, token in httpOnly cookie — never localStorage
- No GraphQL — REST only
- Error boundaries at route level, not component level

## Key Locations
- DB repositories: /lib/repositories/
- API routes: /src/app/api/
- Components: /src/components/
- Types: /src/types/

## DO NOT
- Direct DB queries in route handlers
- JWT in localStorage
- Business logic in React components

MemClaw Workspace

## Current Sprint (Sprint 8, ends 2026-04-04)
- Complete checkout flow (P0)
- Fix Stripe webhook duplicate trigger (P0)

## Decision Log
2026-02-15: Rejected GraphQL — team familiarity, simpler debugging
2026-03-01: Chose Prisma over raw SQL — type safety, migrations

## Known Issues
- Stripe webhook fires twice — check idempotency key
- Redis pool: max 10 connections in dev

## Session History
2026-03-30: Implemented cart calculation, started Stripe integration
2026-03-31: Fixed webhook idempotency, started order confirmation emails

MCP Config

{
  "mcpServers": {
    "memclaw": {
      "command": "npx",
      "args": ["-y", "@memclaw/mcp-server"],
      "env": {
        "MEMCLAW_API_KEY": "your_key",
        "MEMCLAW_WORKSPACE_ID": "your_workspace_id"
      }
    }
  }
}

---

What Each Layer Handles

Context Type	CLAUDE.md	MemClaw
Tech stack	✓
Architecture rules	✓
Current sprint goals		✓
Session history		✓
Decision log		✓
Known issues	Both	Both
Team sharing	Via git	Via workspace

---

Getting Started

1. Create CLAUDE.md in your project root
2. Create a workspace at memclaw.me
3. Configure MCP in .claude/mcp_config.json
4. Create /start and /end slash commands
5. Run your first complete-stack session

Complete your context stack with MemClaw → memclaw.me

---

Written by the Felo Team. We have a business relationship with MemClaw — read our disclosure policy.