deepview

DeepView MCP exposes a single tool named deepview that lets IDEs like Cursor and Windsurf analyze an entire codebase using Gemini's large context window. The server is configured to load a codebase from a single text file (optional) and then answer questions about that codebase. To use it from an IDE, add a new MCP server with the path to the deepview_mcp entry point and your Gemini API key in the environment. The primary usage pattern is: start the server (via the MCP framework in your IDE), then invoke the deepview tool by asking a question about the codebase. You can optionally supply a codebase_file so the server loads it before querying. You can also set the Gemini model via the command-line option, or rely on IDE defaults. Tools available are intentionally minimal: a single tool named deepview that accepts a required question and an optional codebase_file parameter to scope the query to a specific file or dataset.

Prerequisites:

• Python 3.13+ installed on your system
• A Gemini API key from Google AI Studio (or equivalent) with access to the Gemini models you want to use

Installation via pip (recommended):

pip install deepview-mcp

Alternative installation (via Smithery, for Claude Desktop):

npx -y @smithery/cli install @ai-1st/deepview-mcp --client claude

Usage after installation (basic):

deepview-mcp [path/to/codebase.txt]

To configure via your IDE MCP settings, add a server with the following example (adjust paths and keys as needed):

{
  "mcpServers": {
    "deepview": {
      "command": "/path/to/deepview-mcp",
      "args": [],
      "env": {
        "GEMINI_API_KEY": "your_gemini_api_key"
      }
    }
  }
}

If you want to pin a specific Gemini model, pass it through args, for example: --model gemini-2.5-pro-exp-03-25.

Tips:

• Ensure your GEMINI_API_KEY is set in the environment for the MCP to authenticate with Gemini.
• You can preload a codebase by providing a codebase file path in the mcp configuration or by supplying it when querying.
• If you encounter issues with loading the codebase, verify the file path and format expected by your codebase preparation tool (e.g., repomix).
• The server assumes a single-file codebase input; for very large projects consider pre-packaging with repomix as described in the docs.
• For model selection and debugging, you can adjust the --log-level option in the CLI or via your IDE's MCP config where supported.