If you're burning tokens on full file reads and massive grep results, this MCP gives you compressed alternatives. It indexes your codebase and exposes tools like smart_read (outline or focused chunks), smart_search (ranked, limited results), and smart_context (curated multi-file context). The real win is task checkpoints that persist between sessions so you don't re-explain the same context. Ships with workflow metrics showing 85-90% token reduction on debugging and refactoring tasks, though that requires multi-file work where you actually need the index. For single-file edits or when you already know what to touch, native tools are faster. Includes a reactive file watcher, semantic search via local embeddings, and declarative playbooks that chain multiple operations in one call.
claude mcp add --transport stdio arrayo-smart-context-mcp uvx smart-context-mcp