LLVM DevOps Assistant

Inspiration DevOps and repository management are traditionally tied to a screen. Developers constantly monitor commits across massive monorepos like LLVM, reading through raw JSON, SHA hashes, and fragmented commit messages. I wanted to build a hands-free, voice-native interface that allows developers to triage and summarize repository activity using the new Alexa ecosystem and the Model Context Protocol (MCP).

What it does The LLVM DevOps Assistant allows developers to ask Alexa for the latest commits by specific authors. Instead of reading out raw data, the custom backend fetches real-time GitHub data, processes it through AWS Bedrock’s Nova Micro model, and generates a natural-sounding, concise summary optimized for spoken word.

How I built it The backend is built in Node.js and Express. It utilizes: AWS Bedrock (Amazon Nova Micro): Used for fast, low-latency summarization of raw Git commit logs into natural conversational text. Model Context Protocol (MCP) SDK: Implemented to standardize the tool-calling interface for GitHub repository fetching. Alexa Developer Console: Custom VUI intent slots and regex sanitization to handle difficult developer usernames (e.g., stripping spaces from voice inputs). GitHub REST API: To pull real-time commit data sorted by author.

Challenges I ran into (The Friction Log) The biggest engineering hurdle was a direct protocol mismatch between standard Alexa webhooks and the MCP SDK. Alexa expects standard POST webhook responses, but the official MCP Server SDK is heavily built around Server-Sent Events (SSE). To keep the end-to-end voice experience alive without abandoning the MCP standard, I engineered a custom Express middleware adapter to seamlessly bridge the POST-to-SSE gap.

I have fully documented this workaround, along with several other platform beta bugs, in a detailed Friction Log included in the repository for the Amazon platform team.

Accomplishments that I'm proud of Successfully bridging the bleeding-edge Model Context Protocol with the Alexa ecosystem, proving that voice-native developer tools are a viable and powerful interface for the future of DevOps.

What's next for LLVM DevOps Assistant Expanding the MCP tools to include CI/CD pipeline triggers, allowing developers to not just listen to repository summaries, but actively rollback deployments or trigger builds using voice commands via Alexa.