Device Context Protocol

Install the pydcp package from PyPI. Choose the extras matching your hardware connection: serial for USB/UART, mqtt for network-connected devices, ble for Bluetooth Low Energy. The mcp extra is always required.

# Basic install (serial + MCP bridge)
pip install "pydcp[mcp,serial]"

# All transports
pip install "pydcp[mcp,serial,mqtt,ble]"

DCP devices are described by YAML manifests that declare capabilities (intents) with their parameters, types, and access modes. The project ships example manifests in the examples/ directory. You can also use the built-in simulator to test without hardware.

# Inspect an example manifest
dcp inspect examples/lamp_manifest.yaml

# Run with simulator (no hardware needed)
dcp serve examples/lamp_manifest.yaml --simulator

DCP uses scoped capability tokens to control what an LLM agent can do. Generate a secret, then mint tokens with specific capability scopes and optional TTLs.

# Generate a secret (store this securely)
export DCP_SECRET=$(dcp token keygen)

# Mint a token with specific capabilities and 1-hour TTL
dcp token mint --caps lamp.write,lamp.read --ttl 3600

Run dcp serve with the manifest and the appropriate transport flag. DCP translates between the MCP protocol and the compact DCP wire format, bridging your AI assistant to the physical device.

# Serial (USB) connection
dcp serve examples/lamp_manifest.yaml --serial /dev/ttyUSB0

# MQTT connection
dcp serve examples/lamp_manifest.yaml --mqtt broker.lan:1883 --mqtt-prefix dcp/lamp

# BLE connection
dcp serve examples/lamp_manifest.yaml --ble AA:BB:CC:DD:EE:FF

Open claude_desktop_config.json and add a DCP server entry. Use --simulator for testing without hardware, then switch to --serial or --mqtt for production.

{
  "mcpServers": {
    "smart-lamp": {
      "command": "dcp",
      "args": ["serve", "/path/to/lamp_manifest.yaml", "--simulator"]
    }
  }
}