MCP Setup
Connect Claude Desktop, Cursor, and other MCP clients to your OpenClaw instances via the Model Context Protocol.
The Model Context Protocol (MCP) lets you connect AI clients like Claude Desktop and Cursor directly to your OpenClaw instances. Once set up, you can send messages to your OpenClaw agents, check their status, and switch between machines — all from within your AI client.
Prerequisites
- A running OpenClaw instance on Clawdy (create one first)
- An MCP-compatible client (Claude Desktop, Cursor, or any MCP client)
Step 1: Create an API Key
- Go to Dashboard > Settings > MCP API Keys (direct link)
- Click Create API Key
- Give it a descriptive label (e.g. "Claude Desktop")
- Copy the key immediately — it is only shown once
Your key will look like: clawdy_mcp_a1b2c3d4e5f6...
Step 2: Configure Your Client
Claude Desktop
Open your Claude Desktop configuration file:
- macOS:
~/Library/Application Support/Claude/claude_desktop_config.json - Windows:
%APPDATA%\Claude\claude_desktop_config.json
Add the following to the mcpServers section:
{
"mcpServers": {
"clawdy": {
"url": "https://clawdy.app/api/mcp/YOUR_API_KEY/mcp"
}
}
}
Replace YOUR_API_KEY with the key you created in Step 1.
Restart Claude Desktop to apply the configuration.
Cursor
Open Cursor Settings, go to the MCP section, and add a new server:
- Name:
clawdy - Type:
sse - URL:
https://clawdy.app/api/mcp/YOUR_API_KEY/mcp
Other MCP Clients
Any client that supports the MCP SSE transport can connect using the URL:
https://clawdy.app/api/mcp/YOUR_API_KEY/mcp
Available Tools
Once connected, your MCP client will have access to these tools:
| Tool | Description |
|---|---|
openclaw_chat | Send a message to your OpenClaw agent and get a response |
openclaw_status | Check the health of your active OpenClaw instance |
openclaw_list_machines | List all your machines with their status |
openclaw_switch_machine | Switch which machine is active (by slug) |
Example Usage
Ask Claude: "Use OpenClaw to scaffold a new Next.js app with authentication"
Claude will use the openclaw_chat tool to relay your request to your OpenClaw instance and return the response.
Advanced: Stdio Transport (Direct Connection)
For power users who want lower latency, the clawdy-mcp npm package provides a stdio-based MCP server that talks directly to your OpenClaw instances, bypassing the Clawdy API proxy.
Install
npm install -g clawdy-mcp
Configure Claude Desktop
{
"mcpServers": {
"clawdy": {
"command": "clawdy-mcp",
"env": {
"CLAWDY_API_KEY": "YOUR_API_KEY"
}
}
}
}
The stdio server fetches your machine list from the Clawdy API on startup, then communicates directly with your OpenClaw instances over HTTP. This avoids the extra hop through Vercel and can be faster for heavy usage.
Managing API Keys
- You can create multiple API keys for different clients
- Each key shows its prefix and last-used timestamp in the dashboard
- Delete keys you no longer need from Dashboard > Settings > MCP API Keys
- Keys cannot be retrieved after creation — if you lose one, create a new key
Troubleshooting
"Invalid or expired API key" — Double-check the key in your config. Keys are case-sensitive and must include the full clawdy_mcp_ prefix.
"No OpenClaw machine found" — You need at least one running OpenClaw instance. Create one from the dashboard.
Tools not appearing — Restart your MCP client after updating the configuration. In Claude Desktop, fully quit and relaunch the app.
Timeout errors — OpenClaw responses can take up to 2 minutes for complex tasks. If you consistently hit timeouts, check your instance status with openclaw_status.