Capella

Inspiration

I wanted to explore how next-generation voice assistants can go beyond simple reactive commands and act as proactive, cross-domain hubs. Most smart assistants treat developer tools, energy tracking, and family care as disconnected apps. Capella was born to bridge these silos into a unified Model Context Protocol (MCP) framework, bringing intelligent automation to everyday home, family, and dev-ops workflows.

What it does

Capella connects multiple specialized domains into a single assistant interface through FastMCP primitives:

• Developer Workflows: Queries active GitHub repository metrics, open PRs, and pipeline status.

• Eco-Home Analytics: Calculates real-time electricity usage, financial savings, and CO2 offset.

• Family Care Suite: Dispatches scheduled health reminders, medication nudges, and scans calendar milestones for upcoming birthdays and anniversaries with gift suggestions.

• Creative Synthesis: Triggers AI art generation for ambient smart display renders.

• MCP Resources & Prompts: Exposes live system telemetry (config://capella/system-status) and structured prompt templates (daily_executive_briefing).

How we built it

As a solo builder, I designed Capella using a clean, modular architecture:

• Server Architecture: FastMCP and FastAPI running on Python, exposing endpoints via Streamable HTTP (/sse and /messages).

• AI & Fallback Logic: Integrated Google GenAI and Hugging Face Inference API with graceful local fallback handling to ensure reliability.

• Custom Web Simulator: Designed a dark-mode, gradient Web Client Simulator UI using standard HTML5 and JavaScript to test and demonstrate real-time telemetry output across Tools, Resources, and Prompts.

Challenges we ran into

• Environment & Protocol Restrictions: Navigating Cloud Shell's restrictive browser proxy environment caused persistent CORS and 403 Forbidden origin blocks when attempting to run standard MCP Inspector tools.

• Credential & Keychain Constraints: Managing cloud execution without native OS keychains led to secret store fallback issues.

• Solution: I built a custom, self-hosted Web Simulator directly into the FastAPI server wrapper and applied permissive CORS/Host middleware to achieve reliable, isolated testing without external dependency friction.

Accomplishments that we're proud of

• Full Protocol Compliance: Successfully implemented all three core MCP primitives Tools, Resources, and Prompts under official spec standards.

• Solo Execution: Engineered, debugged, and documented the end-to-end backend architecture, modular toolsets, and custom web simulator independently.

• Polished Presentation: Built a clean UI with real-time JSON telemetry output that cleanly demonstrates complex MCP backend execution.

What we learned

• The Power of Open MCP Standards: Implementing the Model Context Protocol showed how seamlessly an assistant can expose structured capabilities without coupling logic directly to the client interface.

• Resilient System Design: Building robust fallback mechanisms for external AI tools is critical when creating reliable assistant agents for real-world execution.

What's next for Capella

• Native Voice Hardware Pairing: Connect Capella directly to physical smart-speaker hardware as gated developer tools open up.

• Live OAuth Integrations: Upgrade simulated calendar and messaging handlers to live Google Calendar and WhatsApp/Twilio API pipelines.

• Expanded Boundary Delegation: Add time-bound user approvals for proactive home purchases and automated tasking.

Built With

• awsservices
• bedrock
• fastapi
• fastmcp
• gemini
• html
• python
• twilio