Claude Code + Novyx

Two tools for Claude Code: novyx-mcp gives your agents persistent memory, and novyx-hygiene keeps your sessions alive across /compact, /clear, and restarts.

Novyx MCP Server — Persistent Memory

Add persistent memory to Claude Code via the MCP protocol. One command to install, 107 tools available immediately.

Install

claude mcp add novyx-memory -- uvx novyx-mcp

Or add to ~/.claude/mcp.json manually:

{
  "mcpServers": {
    "novyx-memory": {
      "command": "uvx",
      "args": ["novyx-mcp"]
    }
  }
}

Local-First Mode (No API Key)

By default, novyx-mcp uses a local SQLite database at ~/.novyx/local.db. No signup, no API key, no network calls. Memory works instantly.

Cloud Mode (API Key)

Set NOVYX_API_KEY to enable cloud sync, audit trails, rollback, and sharing across machines:

{
  "mcpServers": {
    "novyx-memory": {
      "command": "uvx",
      "args": ["novyx-mcp"],
      "env": {
        "NOVYX_API_KEY": "nram_your_key_here"
      }
    }
  }
}

Get a free API key at novyxlabs.com (5,000 memories, no credit card).

Available Tools (64)

Tool	What It Does
remember	Store a memory with optional tags, importance, source
recall	Semantic search across all memories
forget	Delete a memory by ID
list_memories	List memories with filtering
rollback	Undo a memory operation with cryptographic proof
audit	View the audit trail for any memory
add_triple	Add a knowledge graph triple (subject-predicate-object)
query_triples	Query the knowledge graph
link_memories	Create semantic links between memories
create_space	Create a context space for organizing memories
share_space	Share a context space with another user
cortex_run	Run autonomous memory maintenance
cortex_insights	Get AI-generated insights from memory patterns
replay_timeline	View memory timeline for time-travel debugging
replay_snapshot	Point-in-time snapshot of memory state
replay_diff	Compare memory state between two points
replay_lifecycle	Track a single memory's full history
memory_stats	Usage statistics
And more...	107 tools total

Shared Memory Across Agents

When multiple Claude Code agents share the same API key, they share memory. Agent 1 stores context, Agent 2 can recall it — even in different sessions. This is how Blake Heron (Novyx founder) runs 6+ parallel Claude Code agents as a coordinated team.

---

Novyx Hygiene — Automatic Context Persistence

The problem: Context fills up. You /compact or /clear. Now Claude has amnesia. The manual workaround is export → clear → paste → re-establish. Every time.

Novyx Hygiene makes this automatic.

Install

pip install novyx-hygiene
hygiene install

That's it. From now on:

1. Before /compact: Hygiene auto-saves your session (task, decisions, files, git state)
2. After compact/clear/resume: Hygiene injects your context back into Claude automatically
3. On disk: A .claude/hygiene.md file keeps Claude oriented between sessions

No commands to remember. No context to paste. Claude just knows where you left off.

Commands

hygiene install              # Wire up Claude Code hooks (run once)
hygiene install --user       # Install for all projects
hygiene save <task>          # Manually save session state
hygiene resume               # Print most recent session context
hygiene score                # Check context health (A-F grade)
hygiene list                 # List all saved sessions
hygiene config set api_key X # Enable cloud sync (optional)
hygiene uninstall            # Remove hooks

What Gets Saved

Every session snapshot captures:

• Task description — what you're working on
• Key decisions — architectural choices, tradeoffs
• Status — where you left off
• Git state — branch, modified/staged/untracked files, recent commits
• Working directory — so you resume in the right place
• Timestamp — for freshness scoring

Context Health Scoring

$ hygiene score
Context Health: B (80/100)
========================================

Issues:
  - 12 files in flight
  - Changes span 5 top-level directories — possible mixed concerns

Tips:
  - Consider committing completed work before continuing
  - Consider splitting into focused sessions

How Hooks Work

After hygiene install, your .claude/settings.local.json gets:

{
  "hooks": {
    "PreCompact": [
      { "hooks": [{ "type": "command", "command": "hygiene save --auto --quiet", "timeout": 10 }] }
    ],
    "SessionStart": [
      { "matcher": "compact", "hooks": [{ "type": "command", "command": "hygiene inject", "timeout": 5 }] },
      { "matcher": "clear", "hooks": [{ "type": "command", "command": "hygiene inject", "timeout": 5 }] },
      { "matcher": "resume", "hooks": [{ "type": "command", "command": "hygiene inject", "timeout": 5 }] }
    ]
  }
}

• PreCompact runs before compaction, capturing your current state
• SessionStart runs after compact/clear/resume, injecting your last session context
• Hooks are additive — they won't overwrite your existing hooks

Cloud Sync (Optional)

By default, sessions are saved locally to ~/.novyx_hygiene/sessions/. Add a Novyx API key for cloud persistence and cross-machine sync:

pip install novyx-hygiene[novyx]
hygiene config set api_key nram_your_key_here

---

Using Both Together

The best setup uses both:

• novyx-mcp for persistent memory (what your agents know)
• novyx-hygiene for session persistence (where you left off)

# Install MCP server for persistent memory
claude mcp add novyx-memory -- uvx novyx-mcp

# Install Hygiene for session persistence
pip install novyx-hygiene
hygiene install

Memory persists across agents. Sessions persist across compactions. Nothing gets lost.