The data supports this: - METR 2025: experienced devs are 19% slower with AI, despite feeling 20% faster - DORA 2025: 7.2% increase in delivery instability per 25% unstructured AI adoption - IEEE: a requirements error caught in production costs 10-100x more than during specification
Hoofy is an MCP server (Go, single static binary, zero deps) that gives AI assistants:
1. *Persistent memory* — SQLite + FTS5 + knowledge graph. Observations persist across sessions with typed relations and bidirectional traversal.
2. *Adaptive change pipeline* — 4 types × 3 sizes = 12 flow variants. A small bugfix gets 4 stages, a large feature gets 7. Every change starts with a context-check.
3. *Clarity Gate* — for greenfield projects, it analyzes requirements across 8 dimensions and blocks the pipeline until ambiguity drops below threshold. Business rules are extracted using BRG taxonomy and DDD Ubiquitous Language.
Key design decisions: - Tools are storage tools — the AI generates content, Hoofy saves it and advances the state machine - DIP: tools depend on interfaces (config.Store, memory.Store), composition root wires concretions - One file per tool handler (SRP) - CGO_ENABLED=0 for static binaries across all platforms
32 tools total. Works with any MCP-compatible client (Claude Code, Cursor, VS Code Copilot, Gemini CLI). MIT license.
I'm a self-taught developer (started as a NOC operator, now SRE at a large tech company). This is my first open source Go project. Happy to discuss the architecture and design decisions.