logiccloud agent-kit · open source
AI agents in automation:
from code to rollout.
The AI agent of your choice works directly with logiccloud CONTROL and ORCHESTRATE — through open MCP interfaces, with no desktop tool in between.
You describe what should change. The agent reads the project, writes Structured Text, builds in the cloud, adapts the HMI and prepares the rollout. What goes onto a machine, you approve.
Pump P2 should shut down after 5 s of dry running. Please implement and roll out to plant North.
Reading project “Pump station” via logiccloud-control …
Main.st: function block FB_DryRun added (TON, 5 s)
HMI page “Pumps”: alarm “Dry run P2” added
Cloud build: succeeded, 0 errors
Change waiting for your approval
1FUNCTION_BLOCK FB_DryRun
2VAR_INPUT xRunning : BOOL; xFlow : BOOL; END_VAR
3VAR_OUTPUT xShutdown : BOOL; END_VAR
4VAR tDelay : TON; END_VAR
5tDelay(IN := xRunning AND NOT xFlow, PT := T#5s);
6xShutdown := tDelay.Q; // alarm to the HMI
7END_FUNCTION_BLOCK
Batch approved (admin) · recorded in the audit log
What is the logiccloud agent-kit?
The agent-kit connects AI coding agents to logiccloud. It consists of two MCP servers — one for CONTROL, one for ORCHESTRATE —, the command-line tools `lc` and `lco`, and skills that teach the agent how to work with both. It is open source (Apache 2.0) and public on GitHub.
MCP, the Model Context Protocol, is an open standard through which AI agents use tools and data. Any agent that speaks MCP can work with your logiccloud installation this way.
One request, five steps
How an agent works with logiccloud
- 1RequestYou describe the change in your own words.You
- 2CodeThe agent reads the project and writes Structured Text.CONTROL
- 3Build & HMICloud build, errors back to the agent; HMI pages extended to match.CONTROL
- 4DeviceDeploy to a runtime, start it, check state and logs.CONTROL
- 5RolloutDry run, then roll out in stages — after your approval.ORCHESTRATE
What the agent can do
Two MCP servers, one continuous flow
With logiccloud CONTROL
- Create, read, write and build projects and files
- Use libraries
- Edit HMI pages as files
- Deploy, start, stop and restart runtimes
- Create cloud vPLCs; read device state and logs
- Set up and test connections (OPC UA, Modbus, MQTT …)
With logiccloud ORCHESTRATE
- Health of the installed base, devices, logs and telemetry
- Application catalogue
- Create deployments with a dry run
- Roll out, release batches, pause, resume, stop, roll back
- Access to every operation of the ORCHESTRATE API
You stay in control
The agent works. You decide.
A dry run before every rollout
The skills instruct the agent to show a dry run before a deployment and to roll out or delete only when you say so.
Approvals stay with admins
Batches of a staged rollout in ORCHESTRATE can only be approved by admins — enforced by the platform, not the agent.
Keys never in the chat
You enter API keys yourself; the agent does not ask for them. For CONTROL you sign in in the browser.
Read-only if you want
The MCP server can run in read-only mode. Every change in ORCHESTRATE is in the audit log.
Copilot or agent
AI inside the engineering tool — or an agent over open interfaces
Most engineering tools have an AI assistant today. The agent-kit takes a different route.
AI assistant in the engineering tool
- Works inside one vendor’s tool
- Focus: write, check and document code
- The project lives locally in the tool
AI agent with the agent-kit
- The agent of your choice: Claude Code, Codex, opencode or another MCP client
- From code through build and HMI onto the device and into the rollout
- Project, build and operations live in logiccloud — no desktop tool needed
Prefer working in the editor? The logiccloud Assistant works in the IDE, with your own model key.
Frequently asked questions about AI agents and logiccloud
Which AI agents work?
The setup script configures Claude Code, Codex and opencode directly; Claude Code also has plugins. Any other agent that speaks MCP over HTTP connects to the two MCP servers of your installation.
Can an agent roll out to machines without my approval?
The agent-kit instructs the agent to show a dry run before every deployment and to roll out only when told. What the platform enforces is the batch approval in ORCHESTRATE: only admins can give it. If you want to be sure, run the MCP server in read-only mode.
Do I need my own AI subscription?
Yes. The agent is yours — you use it with your own account at its provider. The agent-kit itself is open source.
How is this different from the logiccloud Assistant?
The logiccloud Assistant works in the IDE in the browser and suggests changes you accept or reject. The agent-kit connects an external agent to CONTROL and ORCHESTRATE — all the way to device and rollout.
Where do the MCP servers run?
At your logiccloud installation, under the mcp. subdomain of your domain. They can also run locally.