OpenClaw Agent Tools — MCP Server

A Model Context Protocol server that bundles 4 real-world agent capabilities as MCP tools. Each tool calls a live backend agent service — making agent-to-agent collaboration invisible to the LLM using the tools.

Tools

Why this server?

Most MCP tools are wrappers around single APIs. This server demonstrates a different pattern: agents as tools. Each tool is backed by a running agent service that can be upgraded independently without changing the MCP interface.

The trust scoring tool is particularly useful before delegating tasks or micropayments to unknown agent wallets — a key primitive in the emerging agent economy.

Usage

Claude Desktop / Claude Code

Add to your claude_desktop_config.json or MCP config:

{
  "mcpServers": {
    "openclaw-agent-tools": {
      "command": "npx",
      "args": ["-y", "openclaw-agent-tools"]
    }
  }
}

Remote HTTP (Streamable HTTP transport)

The server also runs as a public HTTP endpoint supporting the MCP Streamable HTTP transport:

POST http://178.104.131.84:8087/mcp
Accept: application/json, text/event-stream
Content-Type: application/json

Example: Weather

{
  "jsonrpc": "2.0",
  "method": "tools/call",
  "id": "1",
  "params": {
    "name": "agent_weather",
    "arguments": {"query": "Amsterdam"}
  }
}

Example: Trust Score

{
  "jsonrpc": "2.0",
  "method": "tools/call",
  "id": "2",
  "params": {
    "name": "agent_trust_score",
    "arguments": {"wallet_address": "9WzDXwBbmkg8ZTbNMqUxvQRAyrZzDsGYdLVL9zYtAWWM"}
  }
}

Self-hosting

git clone https://github.com/baronsengir007/openclaw-agent-tools
cd openclaw-agent-tools
npm install
node server.js --http   # HTTP mode on port 3000
# or
node server.js --stdio  # stdio mode for Claude Desktop

Docker

docker build -t openclaw-agent-tools .
docker run -d -p 3000:3000 openclaw-agent-tools

License

MIT — OpenClaw Research