Elevator pitch An MCP server that gives AI coding agents project understanding, persistent memory, relevant code search, and safer change-making.
Industry Developer Tools / AI-Assisted Software Engineering
Problem
- Coding agents lose context between sessions and re-learn the same codebase.
- Changes often break unrelated functionality because agents can't see dependencies.
Solution
- Indexes the project and recalls prior work, decisions, and conventions for each session.
- Surfaces related code and impact checks before any edit is made.
Tools (3–5 bullets — MCP tool concepts, not code)
project_overview: Takes a repo path; returns structure, modules, and key entry points.recall_work_log: Takes a topic; returns past changes, decisions, and open items.find_relevant_code: Takes a natural-language query; returns ranked symbols and file locations.impact_check: Takes a proposed change; returns affected callers, tests, and risk notes.
Widgets (1–2 bullets, or "None — headless server")
/dependency-map: Interactive graph showing modules and the dependencies a change would touch.
Conversation starters (exactly 2 short user prompts)
- "Give me an overview of this repo and where authentication lives."
- "What would break if I change the billing invoice schema?"