Codebase Onboarding Agent

Codebase Onboarding Agent is a concept-stage project: an MCP-powered agent that helps new developers navigate unfamiliar repositories — answering "where does X happen?", "what calls this function?", and "how do I add a feature like Y?" with grounded references to actual files and lines.

At a glance

• Problem: Onboarding to a large codebase is slow. READMEs go stale; grep doesn't understand intent.
• Approach: Index repository structure and symbols, expose via MCP tools, answer with RAG over code chunks with file/line citations.
• Stack (planned): Model Context Protocol · GitHub API · embedding index · Claude API.
• Status: Concept — design direction set, not yet implemented as a portfolio demo.

Why this matters

MCP is the integration layer — the agent doesn't need custom GitHub OAuth plumbing for every IDE. The hard part is chunking code well (function-scoped, import-aware) and evaluating answer quality on real onboarding questions, not generating plausible-sounding summaries.

Related portfolio work: DocSignal (retrieval evaluation discipline), LeadSignal (structured API orchestration).