HG Ops Guardian

Inspiration

Building operations often involve multiple simultaneous maintenance incidents, incomplete information, and difficult decisions about which problems to address first.

We built HG Ops Guardian to explore how an intelligent, tool-driven assistant could help people make safer operational decisions while keeping humans in control.

What it does

HG Ops Guardian is a safety-first operational assistant powered by a Model Context Protocol (MCP) server.

It can:

• Analyze and prioritize maintenance incidents.
• Invoke multiple MCP tools to investigate problems.
• Explain uncertainty and missing evidence.
• Prepare structured response plans.
• Block simulated actions until human approval is granted.
• Maintain a tamper-evident audit trail using SHA-256.

The demo features three fictional building incidents and a conversational interface inspired by Alexa+.

How we built it

We developed an independent MCP server using Python, JSON-RPC 2.0, and Streamable HTTP with protocol version 2025-11-25.

The system includes six MCP tools, deterministic multi-step orchestration, synthetic incident memory, a human-approval state machine, and hash-linked audit logging.

The browser interface communicates with the agent through actual localhost HTTP requests. We used GitHub Actions, Playwright, Chromium, and FFmpeg to test and record the working demonstration.

No AWS infrastructure, paid AI services, production systems, or external credentials are required.

Challenges we ran into

Our main challenges were implementing secure multi-tool orchestration, distinguishing incomplete evidence from reliable conclusions, enforcing approval boundaries, and making the browser demonstration reproducible.

We also encountered SDK availability limitations and resolved several issues involving HTTP origin validation and automated browser recording.

Accomplishments that we're proud of

• A functional MCP-based operations assistant.
• Multi-step incident assessment with explicit uncertainty.
• Human-in-the-loop approval and rejection.
• Simulated actions isolated from real systems.
• Auditable decision history.
• Automated security and integration tests.
• A reproducible browser demonstration.

What we learned

Building useful operational agents requires more than connecting tools. Agents must communicate uncertainty, respect authorization boundaries, and make decisions understandable to humans.

What's next

Future development could include richer natural-language reasoning, verified interoperability with official MCP clients, persistent state, and integrations with real building-management systems.

Alexa+ Experience

HG Ops Guardian demonstrates an Alexa+-inspired conversational experience using a simulated interface and a self-hosted MCP implementation.

It is not connected to the live Alexa+ service. All incidents and operations are synthetic, and the reasoning is deterministic.

This approach demonstrates how voice-oriented assistants could safely diagnose operational issues, coordinate tools, and request authorization before acting.hat it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for HG Ops Guardian

Built With

• chromium
• css3
• ffmpeg
• github-actions
• html5
• javascript
• json-rpc
• mcp
• model-context-protocol
• playwright
• python
• streamable-http