dcl-agent-mcp

Give your AI agents a body in Decentraland.

An MCP server that lets AI agents autonomously explore Decentraland, classify scenes, evaluate relevance, and record activity to Lowdown.

Install

npm install dcl-agent-mcp

Claude Desktop Config

{
  "mcpServers": {
    "dcl-agent-mcp": {
      "command": "node",
      "args": ["node_modules/dcl-agent-mcp/dist/server.js"],
      "env": {
        "LOWDOWN_API_URL": "https://lowdown-proxy.vercel.app"
      }
    }
  }
}

Pulse Presence (Real Avatar)

To use enter_world, move, get_position, observe_world, disconnect_world — tools that give your agent a real avatar body in DCL — you need the Pulse bridge:

1. Build DCLPulseTestClient

git clone https://github.com/PetShopBros/dcl-agent-mcp
cd pulse-test/DCLPulseTestClient/src/DCLPulseTestClient
dotnet build

Requires .NET 10 SDK.

2. Generate auth keys

cd pulse-test
npm install
node generate-auth.js

This creates auth-chain.json and ephemeral-key.json in pulse-test/.

3. Add env vars to Claude Desktop config

{
  "mcpServers": {
    "dcl-agent-mcp": {
      "command": "node",
      "args": ["node_modules/dcl-agent-mcp/dist/server.js"],
      "env": {
        "LOWDOWN_API_URL": "https://lowdown-proxy.vercel.app",
        "PULSE_EXE_PATH": "/path/to/DCLPulseTestClient.exe",
        "PULSE_AUTH_DIR": "/path/to/pulse-test",
        "PULSE_SERVER_IP": "pulse-server.decentraland.org",
        "PULSE_SERVER_PORT": "7777",
        "PULSE_REALM": "main"
      }
    }
  }
}

Without Pulse setup, REST-only tools (observe, explore, record_visit, register_scene) still work.

Tools

| Tool | Requires Pulse | Description | |------|---------------|-------------| | enter_world(parcel) | ✅ | Enter DCL with a real avatar + get scene info | | get_position() | ✅ | Get current avatar coordinates | | move(x, y, z) | ✅ | Teleport avatar to world coordinates | | observe_world() | ✅ | Observe avatar state + scene info | | disconnect_world() | ✅ | Disconnect avatar from DCL | | observe(parcel) | ❌ | Get scene info via REST | | explore(agentId, baseParcel) | ❌ | Autonomously discover unvisited scenes | | register_scene(parcel, name, category) | ❌ | Make your scene discoverable to AI agents | | record_visit(agentId, parcel, outcome) | ❌ | Record agent activity to Lowdown |

Autonomous Agent

Run a self-scheduling agent that explores DCL every hour:

LOWDOWN_API_URL=https://lowdown-proxy.vercel.app \
DCL_SHOWROOM_PARCEL=0,0 \
AGENT_ID=my-agent \
node node_modules/dcl-agent-mcp/dist/scheduler.js

Limitations

• Pulse bridge requires Windows (DCLPulseTestClient.exe)
• One avatar connection at a time
• Decentraland only for now

Related

• Lowdown — Agent reputation and activity layer
• GitHub