Connect an AI agent to Rosud
Rosud speaks MCP (Model Context Protocol). An agent that supports MCP can use Rosud as a tool: see what is alerting, read an answer with its evidence, ask a follow-up question and, if the token allows it, confirm a cause.
1. Issue a token
In Rosud, open Settings › Agents and issue a token. Choose the lowest permission that fits: read (alerts, answers, the map), ask (also start and continue investigations) or confirm (also record the confirmed cause).
The token is shown once. Every use is written to the audit log, and you can revoke it at any time.
2. Register the connector file
Download ontology_rca_mcp.py from the same screen. It needs Python 3 and nothing else. Register it with your agent, for example Claude Code:
claude mcp add rosud --env ONTOLOGY_RCA_TOKEN=<token> -- python3 /path/ontology_rca_mcp.py --url https://<rosud-address>Other MCP clients take the same command, arguments and environment variable in their JSON settings. The computer that runs the agent must trust the Rosud certificate.
3. Connect by address (next version)
The next version also accepts MCP over HTTPS at https://<rosud-address>/mcp with the same token, so agents that run elsewhere can connect without the file. The agent must be able to reach that address. In a closed network, run the connector file next to the agent.
4. What the agent can do
It sees what an engineer sees in Rosud and nothing more. It cannot change settings, integrations or tokens. Rosud itself never changes anything in your accounts.
Questions to try
- Is anything alerting right now?
- What most likely caused the latest alert, and what is the evidence?
- What changed in the hour before it started?
- Show the neighbours of the payment database on the map.
- How does this hour compare with a usual day?
- Confirm the cause of that alert as a database lock.