Summary

A second model as a tool, with search already attached.

The value is not the wrapper but what it wraps: a Gemini call that has Google Search and URL reading turned on, so one question can pull in pages your main assistant never fetched. It is one tool with one required parameter, and the API bill is yours.

What it is

A Deno server exposing a single tool. The prompt goes to Gemini with the `googleSearch` and `urlContext` tools enabled, so the model can look things up and read any URLs you include in the text before it answers.

What you get

One answer to a free-text question, with the web searched automatically for current information — `ask_gemini`
URLs written into the prompt read and analysed as part of that same call, with no separate fetch step
Which model answers, set by you through `GEMINI_MODEL`

Requirements

A Google Gemini API key in `GEMINI_API_KEY` and a model name in `GEMINI_MODEL`. Deno v1.40+. Runs straight from JSR: the client launches `deno` with `--allow-net`, `--allow-env` and `jsr:@cong/gemini-mcp`. Both variables must be set in the same shell that launches the server — the project calls this out because it is the usual cause of a failed start.

Setup effort

Build from source — clone the repository and build it, then point your client at the binary