MCP-based Network Configuration Agent

A Model Context Protocol server exposing secure, capability-limited tools so agents can read and reason about network configuration data.

AI Agents and MCP, professional work for an employer

Problem and motivation

Configuration data lives behind several services, and general-purpose LLMs lack a secure, standardized mechanism to inspect configuration schemas and safely query device state.

Solution and architecture

Configuration capabilities are exposed as MCP tools with typed, constrained inputs, so an agent can discover and call them in a controlled way — querying device configurations, validating syntax, and simulating provisioning changes without touching production parameters.

My role

Design and implementation of the agent's tool interface and prompts. This was professional work for an employer.

Technology stack

Spring AI
MCP
Java
Spring Boot
REST APIs

Key engineering decisions

Expose capabilities as small, well-described tools and validate tool arguments on the server.
Enforce strict sandboxing of tool execution to prevent unauthorized modification of production parameters.
Design clear JSON-RPC interfaces for seamless integration with MCP-compatible clients.

Lessons learned

Standardizing context providers through MCP dramatically simplifies building secure agentic workflows.
Tool schemas are part of the product interface; precise descriptions reduce ambiguity.

AI-powered Telecom Operations Assistant
Telecom and Network Automation, professional work
An agent-based incident triage and alarm correlation assistant that helps network operations teams work with alarms, incidents, and service impact through natural-language requests.
- Java
- Spring Boot
- Spring AI
- LLMs
- Tool calling
- Structured outputs
- MySQL