Your IDE assistant works from the files it happened to open; this one answers with the whole repository in front of it. The cost is that the snapshot is only as current as the last time you regenerated the file.
An MCP server that lets Cursor, Windsurf and other MCP-capable IDEs put a question to Gemini with a whole codebase in context. You supply the codebase as a single file; the server sends it along with the question.
- One tool, `deepview`: a required `question` about the codebase, and an optional `codebase_file` to load before answering
- The model chosen at launch with `--model` — the default is `gemini-2.0-flash-lite`
- The codebase file pinned in the MCP config so every question hits the same snapshot, or supplied per call when you work across several projects
- Log level controlled with `--log-level`, from DEBUG through CRITICAL
Python 3.13+ and a Gemini API key from Google AI Studio, supplied as `GEMINI_API_KEY`. Install with `pip install deepview-mcp`. Your codebase has to arrive as one file — the author points at repomix to produce it, with an include and exclude list so vendored directories stay out. MIT.
One command plus a key — pip install deepview-mcp, then supply credentials