Proton MCP
Proton MCP is Proton's connector for AI assistants. With it, Claude, ChatGPT, Claude Code, or Codex use Proton on your behalf: you ask in natural language, and the assistant looks things up, builds, and runs automations without you having to open Proton.
The connector follows the Model Context Protocol (MCP), the open standard AI assistants use to talk to other systems.
What you can do
Everything the connector does goes through the same rules as the screens: your profile's permissions, the automations' team rule, and the organization's plan.
- Automations and data: create and change automations, components, parameters, datasets, and the values of each step, including in batches; set up environments and their variables.
- Runs: set up groups and cycles, run now or schedule, create recurring runs, track the status, stop runs, and turn cloud runners on and off.
- Investigate failures: read a run's log, error, AI diagnosis, and evidence, search for text in the logs of many runs, and find a run by its Correlation ID.
- Schedule and reports: what runs today or this week, what didn't run, the automations that fail the most, one period's pass rate against another's, and the health of your automation assets.
- Processes and traceability: business processes, the BPMN diagram, the automation document, the traceability matrix, the impact analysis of a change, and the change history.
- Jira, Xray, and Zephyr: link requirements, open bugs from a run, and publish results to Test Plans.
- Robot operations: the robots' exception queue, with assignment, comments, and reruns.
- Project from scratch: start an automation project in Python or Java with the code already in the Proton structure, and check the code and the Proton setup before running. See Project from scratch.
The full list is in Tools, and there are ready-made requests in Example requests.
How it works
- You add the Proton connector to your assistant, using the connector URL.
- The first time, the assistant opens Proton in your browser. You sign in with your Proton account, choose the organization, and authorize.
- From then on, when a request involves Proton, the assistant calls the connector's tools. Each tool does just one thing, such as listing automations or running a dataset, and uses the same paths as the Proton screens.
The assistant acts as you: it only sees and changes what you would see and change on screen, and whatever it saves is recorded under your name.
The AI assistant inside Proton uses the same tools.
Before you start
-
Connector enabled in the organization. The connector is off by default. Atomic enables it per organization, either read-only or read and write. If the authorization screen says none of your organizations has the connector enabled, talk to whoever manages Proton at your company.
-
Full access license. Users with a Runner-only license can't use the connector.
-
A supported assistant:
- Claude (claude.ai, the desktop app, and the mobile app);
- Claude Code;
- ChatGPT;
- Codex.
Other MCP clients aren't supported yet.
URL
| Environment | Connector URL |
|---|---|
| Proton (production) | https://app.protoncloud.com.br/api/mcp |
Your environment's URL also appears in Proton, under My Profile > AI Connections, with the Copy URL button.
Next step: Connect the assistant.