Case 02APPLIED AI
Enterprise AI Agent Tooling
The tools behind an agency-wide LLM platform at the U.S. Department of Defense: MCP servers that let its assistant do real work, an air-gapped IDE for AI coding agents, and the skills and training that help teams build with them.
- Role
- AI Agents Software Engineer
- Timeline
- 2026 – Present
- Status
- Live and growing
- 3 of 4
- Production MCP servers on the platform
- 1,000s
- Tool calls handled every month
- 15
- Agent-tooling repositories I review
This work runs on a closed network, so there are no screenshots.
01 The mission
An enterprise AI assistant is only as useful as the tools it can call. Teams needed it to search internal knowledge, draw charts, and help write code, on a network with no internet access, where every tool had to be secure, reviewable, and easy for other teams to build on.
02 The tool belt
Giving an assistant real capabilities.
MCP servers in production
I built 3 of the platform's 4 production MCP servers in Python with FastMCP, handling thousands of tool calls a month. They include a multi-step research agent that returns one cited answer from many sources, and schema-validated tools that turn plain English into charts and diagrams in seconds.
An IDE for agents, fully offline
I shipped an air-gapped Electron and TypeScript IDE for AI coding agents on Linux and Windows, with offline package mirrors, a model gateway, PKI authentication, and in-app updates. Dozens of employees were using it before it was ever advertised.
Raising the bar across teams
I own code review for the platform's 15 agent-tooling repositories, more than 20 merge requests a week, and wrote the agent skills and onboarding course teams use to build with AI.
03 Built with
- Python
- FastMCP
- Model Context Protocol
- TypeScript
- Electron
- Svelte