# OpenClaw Integration Guide
This guide covers integrating MoltBrain with [OpenClaw](https://github.com/openclaw/openclaw), the popular personal AI assistant.
## Overview
MoltBrain provides three integration methods for OpenClaw:
1. **Extension**: Full integration with OpenClaw's extension system
2. **Skill**: Lightweight tool-based integration
3. **MCP Server**: Protocol-based integration for advanced setups
## Quick Start
### Extension (Recommended)
#### For Bundled Installation
If you're using OpenClaw from source or have cloned the repository:
```bash
# Clone into OpenClaw extensions directory
cd your-openclaw-installation/extensions
git clone https://github.com/nhevers/moltbrain.git moltbrain
# Build the extension
cd moltbrain/integrations/openclaw
npm install && npm run build
```
**Important:** Enable the plugin (bundled plugins are disabled by default):
```bash
pnpm openclaw plugins enable moltbrain
```
Add to OpenClaw config:
```json
{
"plugins": {
"entries": {
"moltbrain": {
"enabled": true,
"config": {
"dataDir": ".moltbrain",
"maxMemories": 10,
"autoCapture": true
}
}
}
}
}
```
#### For Standalone Installation
```bash
# Clone into user extensions directory
cd ~/.openclaw/extensions
git clone https://github.com/nhevers/moltbrain.git moltbrain
# Build the extension
cd moltbrain/integrations/openclaw
npm install && npm run build
# Enable the plugin
pnpm openclaw plugins enable moltbrain
```
Restart OpenClaw gateway and you're done!
## How It Works
### Memory Flow
```
User Message → OpenClaw → MoltBrain hooks
↓
Search relevant memories
↓
Inject into context
↓
Generate response
↓
Extract observations
↓
Save to memory store
```
### Automatic Capture
Claude Recall automatically captures:
| Type | Trigger | Example |
|------|---------|---------|
| Preference | "I prefer...", "I like..." | "I prefer tabs over spaces" |
| Decision | "Let's...", "We should..." | "Let's use PostgreSQL" |
| Learning | Technical explanations | "The API uses OAuth2" |
| Context | Project-specific info | File paths, configs |
### Context Injection
Before each response, claude-recall searches for relevant memories and injects them:
```xml
<recalled_context>
- [preference] User prefers TypeScript over JavaScript
- [decision] Using PostgreSQL for the database
- [learning] The auth module uses JWT tokens
</recalled_context>
```
## Configuration
### Extension Config
```json
{
"plugins": {
"entries": {
"claude-recall": {
"enabled": true,
"config": {
"dataDir": ".claude-recall",
"maxMemories": 10,
"autoCapture": true
}
}
}
}
}
```
**Note:** The plugin must be explicitly enabled using `pnpm moltbot plugins enable claude-recall` if it's installed as a bundled extension (in MoltBot's `extensions/` directory).
### Skill Config
```json
{
"skills": {
"moltbrain": {
"maxMemories": 10,
"channels": ["discord", "slack"]
}
}
}
```
### MCP Config
```json
{
"mcp": {
"servers": {
"moltbrain": {
"command": "node",
"args": ["~/.openclaw/extensions/moltbrain/src/mcp/server.js", "--stdio"]
}
}
}
}
```
## Channel-Specific Setup
### Discord
No additional setup needed. Memories are captured from all Discord interactions.
### Slack
No additional setup needed. Works with both DMs and channels.
### iMessage
Memories are stored per-conversation. Group chats are supported.
### Microsoft Teams
Works with both personal and team channels.
## Advanced Usage
### Manual Memory Management
Use the skill tools to manually manage memories:
```
# Save important info
save_memory("The production server is at api.example.com", type="context")
# Search memories
search_memories("server configuration")
# Get relevant context
recall_context("deploying to production")
```
### Filtering by Channel
Limit memory capture to specific channels:
```json
{
"channels": ["discord", "slack"]
}
```
### Memory Types
| Type | Use Case |
|------|----------|
| `preference` | User preferences and settings |
| `decision` | Architectural and design decisions |
| `learning` | Technical knowledge and insights |
| `context` | Project-specific information |
## Troubleshooting
### Plugin Not Loading
**Problem:** Plugin shows as "disabled" in `pnpm moltbot plugins list`
**Solution:**
1. Enable the plugin: `pnpm moltbot plugins enable claude-recall`
2. Restart the MoltBot gateway
3. Verify in config: `plugins.entries.claude-recall.enabled` should be `true`
### Tools Not Appearing
**Problem:** Tools (`recall_context`, `search_memories`, `save_memory`) don't appear in agent's tool list
**Solution:**
1. Verify plugin is enabled (see above)
2. Rebuild TypeScript: `npm run build` or `pnpm tsc`
3. Restart MoltBot gateway
4. Check logs for registration messages: `[claude-recall] Extension register() called`
### Memories Not Being Captured
1. Check `autoCapture` is `true` in config
2. Verify the channel is enabled
3. Check logs for errors
4. Ensure plugin is enabled and loaded
### Context Not Being Injected
1. Verify memories exist: Use `search_memories` tool
2. Check `maxMemories` setting
3. Ensure extension is loaded and enabled
4. Check that tools are available to the agent
### MCP Connection Issues
1. Verify server is running: `npm run mcp:status`
2. Check port availability (default: 3847)
3. Review MCP logs
### Bundled Plugin Disabled by Default
**Note:** If the extension is installed in MoltBot's `extensions/` directory (bundled), it will be disabled by default. You must explicitly enable it using `pnpm moltbot plugins enable claude-recall`.
## Performance
- Memory search: ~10ms average
- Context injection: ~5ms overhead
- Storage: ~1KB per memory entry
## Security
- All data stored locally
- No external API calls
- Memories are per-installation
