Agent

The Magistrala IoT Agent runs in the project Docker environment and acts as a bridge between local services and the Magistrala platform. Inside the container stack it loads runtime configuration, connects to MQTT, subscribes to remote commands, exposes a local HTTP API, manages Node-RED flows, executes shell commands, maintains terminal sessions, and tracks heartbeat messages from local services.

Configuration

The agent can start from a local config.toml file or bootstrap itself from Magistrala using an external ID and key. Environment variables are merged with file-based configuration at startup.

How Agent Works

Provisioning comes first

Before the agent container is started, the project provisions the Magistrala resources the agent depends on. Running make provision or make run_provision executes scripts/provision.sh, which creates the required cloud-side resources and writes the resulting runtime values into configs/config.toml.

The provisioning flow creates or configures:

• a Magistrala client for the agent
• a channel used for request and response messaging
• a bootstrap configuration for the device
• a Rule Engine rule used by the example setup

By default the provisioning script creates the agent client with the name agent-mock-device, so the MQTT examples in this project use that same provisioned client name.

The provisioning script supports both local and cloud targets:

• Local Docker target: if MG_API is not set, provisioning uses the local service defaults such as http://localhost:9006, http://localhost:9005, http://localhost:9008, http://localhost:9013, and the MQTT default ssl://host.docker.internal:8883
• Cloud target: if MG_API=https://cloud.magistrala.absmach.eu/api is set, the provisioning requests are sent to Magistrala Cloud and the default MQTT target becomes ssl://messaging.magistrala.absmach.eu:8883

This is why provisioning is the first step in the Docker workflow: the agent container expects its channel, MQTT credentials, workspace information, and bootstrap data to exist before it starts.

Startup flow

At startup the agent:

1. Starts inside the Docker Compose stack.
2. Loads environment-based defaults.
3. Reads config.toml when available.
4. Optionally fetches bootstrap configuration from Magistrala.
5. Merges the provisioned values into its runtime config.
6. Validates required runtime values such as domain_id, channels.id, mqtt.url, mqtt.username, mqtt.password, and server.broker_url.
7. Connects to FluxMQ for heartbeat subscriptions.
8. Connects to MQTT for remote request and response messaging.
9. Starts the local HTTP server.

Messaging flow

The agent subscribes to:

• m/<workspace_id>/c/<channel_id>/req for incoming remote commands.
• heartbeat.<service-name>.<service-type> through the local pub/sub broker for service heartbeat tracking.

The agent publishes responses to:

• m/<workspace_id>/c/<channel_id>/res for command results and status messages.
• terminal response topics while an interactive terminal session is active.

All remote MQTT commands use SenML JSON array payloads. The n field selects the subsystem:

Core responsibilities

Node-RED behavior

The agent supports these Node-RED commands:

When flows are sent through nodered-deploy or nodered-add-flow, the payload is base64-decoded, normalized, and then forwarded to Node-RED. For imported single-flow payloads, the agent rekeys Node-RED IDs to avoid duplicate ID conflicts. The agent also patches MQTT-related flow values so deployed flows use the runtime device/channel context.

Deployment

The documented flow for this project is container-based.

1. Provision Magistrala resources

Set the required Magistrala values and run provisioning:

export MG_PAT=<personal-access-token>
export MG_DOMAIN_ID=<domain-id>
make provision

This step creates the required Magistrala client, channel, bootstrap config, and example rule, then updates configs/config.toml with the provisioned values used by the container setup.

For a cloud deployment target, set the cloud API base before provisioning:

export MG_API=https://cloud.magistrala.absmach.eu/api
export MG_PAT=<personal-access-token>
export MG_DOMAIN_ID=<domain-id>
make provision

When MG_API points to https://cloud.magistrala.absmach.eu/api, the provisioning flow targets Magistrala Cloud. In that mode, MG_AGENT_MQTT_URL=ssl://messaging.magistrala.absmach.eu:8883 also points the agent at the cloud MQTT broker instead of the local Docker-accessible one.

2. Build the Docker images

git clone https://github.com/absmach/agent
cd agent

make all && make dockers_dev

3. Start the Docker stack

make run

This starts the agent together with the supporting containers such as Node-RED, the Agent UI, and the local messaging components used by the development setup.

4. Stop the stack

make stop
make clean_volumes

Agent UI

The Docker stack also starts a web-based management UI at http://localhost:3002.

The UI is used for local interaction with the running agent container and provides:

• Configuration to view and save agent settings such as server, channels, mqtt, nodered, and log
• Node-RED tools to ping Node-RED, inspect state, fetch flows, deploy flows, and add a single flow tab from a local JSON file
• Services to view heartbeat-discovered local services
• Execute Command to run shell commands and inspect terminal-style output

The UI is built with Elm and served as part of the Docker environment, so it is part of the normal development workflow rather than a separate optional component.

HTTP API

Base URL defaults to http://localhost:9999.

Execute a shell command

curl -X POST http://localhost:9999/exec \
  -H "Content-Type: application/json" \
  -d '{
    "bn": "req-1:",
    "n": "exec",
    "vs": "pwd"
  }'

View current config

curl http://localhost:9999/config

Deploy Node-RED flows

FLOWS=$(base64 -w 0 examples/nodered/speed-flow.json)

curl -X POST http://localhost:9999/nodered \
  -H "Content-Type: application/json" \
  -d "{\"command\":\"nodered-deploy\",\"flows\":\"$FLOWS\"}"

Add a flow tab without replacing existing flows

FLOWS=$(base64 -w 0 examples/nodered/modbus-flow.json)

curl -X POST http://localhost:9999/nodered \
  -H "Content-Type: application/json" \
  -d "{\"command\":\"nodered-add-flow\",\"flows\":\"$FLOWS\"}"

Ping Node-RED

curl -X POST http://localhost:9999/nodered \
  -H "Content-Type: application/json" \
  -d '{"command":"nodered-ping"}'

View Node-RED runtime state

curl -X POST http://localhost:9999/nodered \
  -H "Content-Type: application/json" \
  -d '{"command":"nodered-state"}'

Fetch current flows

curl -X POST http://localhost:9999/nodered \
  -H "Content-Type: application/json" \
  -d '{"command":"nodered-flows"}'

View registered services

curl http://localhost:9999/services

MQTT Commands

Remote control happens over MQTT using SenML payloads published to m/<workspace_id>/c/<channel_id>/req.

Execute a command

mosquitto_pub \
  -h <mqtt-host> -p 8883 --capath /etc/ssl/certs \
  -u <client-id> -P <client-secret> \
  -t "m/<workspace_id>/c/<channel_id>/req" \
  -m '[{"bn":"req-1:","n":"exec","vs":"ls,-la"}]'

View config

mosquitto_pub \
  -h <mqtt-host> -p 8883 --capath /etc/ssl/certs \
  -u <client-id> -P <client-secret> \
  -t "m/<workspace_id>/c/<channel_id>/req" \
  -m '[{"bn":"req-1:","n":"config","vs":"view"}]'

Deploy Node-RED flows

FLOWS=$(base64 -w 0 examples/nodered/speed-flow.json)

mosquitto_pub \
  -h <mqtt-host> -p 8883 --capath /etc/ssl/certs \
  -I "agent-mock-device" \
  -u <client-id> -P <client-secret> \
  -t "m/<workspace_id>/c/<channel_id>/req" \
  -m "[{\"bn\":\"req-1:\",\"n\":\"nodered\",\"vs\":\"nodered-deploy,$FLOWS\"}]"

Add a flow tab without replacing existing flows

FLOWS=$(base64 -w 0 examples/nodered/modbus-flow.json)

mosquitto_pub \
  -h <mqtt-host> -p 8883 --capath /etc/ssl/certs \
  -I "agent-mock-device" \
  -u <client-id> -P <client-secret> \
  -t "m/<workspace_id>/c/<channel_id>/req" \
  -m "[{\"bn\":\"req-1:\",\"n\":\"nodered\",\"vs\":\"nodered-add-flow,$FLOWS\"}]"

Ping Node-RED

mosquitto_pub \
  -h <mqtt-host> -p 8883 --capath /etc/ssl/certs \
  -I "agent-mock-device" \
  -u <client-id> -P <client-secret> \
  -t "m/<workspace_id>/c/<channel_id>/req" \
  -m '[{"bn":"req-2:","n":"nodered","vs":"nodered-ping"}]'

View Node-RED runtime state

mosquitto_pub \
  -h <mqtt-host> -p 8883 --capath /etc/ssl/certs \
  -I "agent-mock-device" \
  -u <client-id> -P <client-secret> \
  -t "m/<workspace_id>/c/<channel_id>/req" \
  -m '[{"bn":"req-3:","n":"nodered","vs":"nodered-state"}]'

Fetch current flows

mosquitto_pub \
  -h <mqtt-host> -p 8883 --capath /etc/ssl/certs \
  -I "agent-mock-device" \
  -u <client-id> -P <client-secret> \
  -t "m/<workspace_id>/c/<channel_id>/req" \
  -m '[{"bn":"req-4:","n":"nodered","vs":"nodered-flows"}]'

Best Practices

• Run provisioning before starting the Docker stack so the agent container has valid client, channel, and bootstrap data.
• Keep channels.id, domain_id, and MQTT credentials aligned, because the agent depends on them to subscribe and publish correctly.
• Route all Node-RED flow deployment through the agent when flow payloads need runtime patching.
• Use heartbeat publishing for local services that should appear in the agent UI and /services API.
• Keep terminal session timeouts short on production devices.
• Restrict shell command access to trusted operators only.