Telecom

Domain: Telecom

We construct a simulated telecom customer support environment that combines several enterprise systems commonly deployed by real-world operators. The environment includes a customer database, an identity and access management layer for account authentication, billing and revenue-management records, customer support ticketing, communication logs across email and SMS, order management, a network-operations-style outage tracker, field-service scheduling, and a customer community forum. This multi-system design reflects the architecture of operational telecom platforms, where agents must reason jointly over customer identity, account access, billing state, service requests, network availability, and external-facing communication channels.

The environment is backed by structured backend tables containing customer profiles, account records, billing entries, support tickets, communication histories, service orders, outage reports, appointment schedules, and community forum activity. The agent has no direct access to the underlying storage and must interact with the environment exclusively through MCP tool calls. This tool-only interface both improves experimental control and more faithfully models the mediated interaction patterns of real enterprise support stacks.

The agent's action space consists of multiple MCP tools organized into 9 functional categories (the MCP-tool table): customer profile management, account and authentication, billing, support ticketing, communication logs, order management, network operations, scheduling, and community forum operations. Concretely, these tools allow the agent to retrieve and update subscriber records, manage account credentials, inspect and modify bills, create and track support tickets, review historical email and SMS interactions, manage service or product orders, query outage conditions and restoration timelines, schedule customer-linked appointments, and access forum threads and posts. By spanning both private operational systems and customer-facing interaction channels, the telecom environment presents a realistic cross-system action space for studying agent capability and security.