@codebase@codebase isCursor’s @codebase feature indexes your repository and injects relevant chunks into the model prompt when you reference it in a chat. Indexing is automatic and tied to the Cursor IDE.
| Dimension | Cursor @codebase |
NeuralMind |
|---|---|---|
| Host | Cursor IDE only | Any agent (Claude Code, Cursor, Cline, Continue, Claude Desktop, CLI) via MCP |
| Index unit | File chunks | Graph nodes (functions, classes) + communities + rationales |
| Retrieval | Chunk similarity | 4-layer progressive disclosure (identity → summary → clusters → search) |
| Output | Raw chunks into prompt | Structured, token-budgeted context with reduction metrics |
| Tool-output compression | None | PostToolUse hooks compress Read/Bash/Grep results |
| Offline | No (Cursor cloud) | Yes — nothing leaves your machine |
| Cost | Bundled in Cursor subscription | Free, local |
| Install methods | Paid Mac/Windows/Linux IDE installer | pip / pipx / uv / Docker / source — runs anywhere Python does |
| Adapts to you | No | Yes — a Hebbian synapse layer learns associations from your usage automatically (co-activation with decay), no manual step |
@codebase if you only use Cursor and are happy with the built-in behavior.Read/Bash/Grep call — not just explicit @codebase invocations.They are not mutually exclusive: run NeuralMind’s MCP server inside Cursor and you get compression on top of Cursor’s own indexing.