MCP Integrations

Your calendar, your task system, your design files, your inbox, your code host. Any service that exposes an MCP server can become a context source or an action surface for the Parabellum OS agent mesh. The operator approves each connection. The operator can revoke it at any time.

Each MCP server is registered through the Command Interface. On registration, Parabellum OS reads the server's tool manifest, surfaces the available tools to the operator, and requests explicit permission before any tool is called. Permissions are scoped per server and per tool.

• Google Calendar. Read schedule, draft events.
• Gmail. Read threads, draft replies, create labels.
• Google Drive. Search and read files, draft new documents.
• Slack. Read channels the operator has joined, draft messages.
• GitHub. Read repositories, draft issues and pull requests.

Any MCP-compliant server can be added through the Command Interface by providing the server URL and a connection name. The system reads the manifest, requests scoped permissions, and registers the tool surface. Custom servers run with the same permission model as built-in ones.

Three states per tool: not granted, granted for this session, granted persistently. Persistent grants are revocable from the operator profile. No tool runs without an active grant.

Every MCP tool call is logged with the agent that initiated it, the tool name, the input, and the result. The log is queryable through the API and viewable in the Command Interface.

MCP is not a backdoor into your accounts. It is not a way for agents to act without your knowledge. It is a permissioned, audited bridge between Parabellum OS and the services you choose to connect.