openclaw-mcp
by Codename-11
README.md
# openclaw-mcp
MCP server for Claude Code to talk to OpenClaw AI agents (Daemon, Soren, Ash, etc.) via the gateway API.
## What it does
Gives Claude Code three tools:
- **`ask_agent`** — Send a message to any OpenClaw agent and get their response inline
- **`list_agents`** — Discover all available agents on the gateway
- **`agent_status`** — Check if a specific agent is active
## Quick Setup
```bash
# Install globally
npm install -g openclaw-mcp
# Add to Claude Code (simplest)
claude mcp add openclaw -- npx openclaw-mcp
# Or with explicit config
claude mcp add openclaw \
--env OPENCLAW_GATEWAY_URL=http://172.16.24.250:18789 \
--env OPENCLAW_GATEWAY_TOKEN=your_token \
-- npx openclaw-mcp
```
## Manual Setup
Add to your `.claude/settings.json` (or `~/.claude.json` for global):
```json
{
"mcpServers": {
"openclaw": {
"command": "npx",
"args": ["openclaw-mcp"],
"env": {
"OPENCLAW_GATEWAY_URL": "http://172.16.24.250:18789",
"OPENCLAW_GATEWAY_TOKEN": "your_token"
}
}
}
}
```
## Configuration
| Variable | Default | Description |
|---|---|---|
| `OPENCLAW_GATEWAY_URL` | `http://localhost:18789` | OpenClaw gateway URL |
| `OPENCLAW_GATEWAY_TOKEN` | _(required)_ | Auth token from your OpenClaw config |
Find your token in `~/.openclaw/openclaw.json` under `gateway.token`.
## Usage in Claude Code
Once configured, just talk to Claude Code naturally:
- "Ask Daemon about the ClawPort architecture"
- "Check which agents are available"
- "Ask Soren to review this code approach"
- "What's the status of the Ash agent?"
## Tools Reference
### `ask_agent`
Sends a message to an OpenClaw agent and returns their response.
```
agent: "daemon" | "soren" | "ash" | "mira" | "jace" | "pip" | ...
message: "Your question or request"
```
Uses session key `agent:{name}:mcp` so MCP conversations are isolated from Discord/ClawPort sessions.
### `list_agents`
No params. Returns all models/agents registered on the gateway.
### `agent_status`
```
agent: "ash"
```
Returns the agent's name, model ID, and availability status.
## Local Development
```bash
git clone https://github.com/Codename-11/openclaw-mcp
cd openclaw-mcp
npm install
npm run build
# Test
OPENCLAW_GATEWAY_URL=http://localhost:18789 \
OPENCLAW_GATEWAY_TOKEN=your_token \
node dist/index.js
```
## License
MIT
TDQS
A4/5.0
Scored across 3 tools
Disambiguation5/5
Each tool serves a clearly distinct purpose: listing agents, checking a specific agent's status, and sending a message. There is no overlap between these operations.
Naming Consistency4/5
Most tools follow a verb_noun pattern (ask_agent, list_agents), but agent_status deviates by using noun_noun. The inconsistency is minor and the names remain clear.
Tool Count5/5
Three tools is a well-scoped set for an agent interaction server, covering the core needs of discovery, status checking, and messaging without unnecessary bloat.
Completeness4/5
The set covers the primary lifecycle of interacting with agents, but lacks explicit session management (e.g., start/stop). This is a minor gap for a basic agent gateway server.
Maintenance
ActivityInactive
ResponsivenessNo issues