call_google_search
Takes a prompt and generates an answer from Gemini using Google Search grounding
LABSCO SUMMARY
Written for Gemini CLI, and the README says so.
The author's own notes are the most useful part of the README: Cursor never managed to call the tools, and the client they got working was Gemini CLI on the Windows side rather than inside WSL. No Gemini CLI settings are published, so use the Docker configuration on this page in any client that launches stdio servers. Ignore the README's tool section: its four get_connpass_* entries are left over from the connpass server this began as. The code registers call_google_search, call_gemini_or_vertex_ai and get_random_greeting.
What it is
A Deno MCP server that puts Gemini's Google-search grounding behind a tool call. It runs from a local checkout or a Docker image, and the Deno permission flags in the documented launch restrict it to a single outbound host.
What you get
- Gemini answers grounded in Google search —
call_google_search— plus a plain Gemini or Vertex AI call,call_gemini_or_vertex_ai - A pinned network permission on the launch line:
--allow-net=generativelanguage.googleapis.com, so the process can reach that host and nothing else - Three runtimes for the same server — Deno directly, a Node build via
npm ciandnpm run build, or Docker - Repeatable checks:
deno task test:tokyo,deno task test:google_searchanddeno task test:google_search_simple, plus a one-linetools/listprobe you can pipe into the server
Requirements
A Gemini API key as GEMINI_API_KEY, supplied through an --env-file or -e on the Docker run. Deno is the recommended runtime, and the client entry passes --allow-net=generativelanguage.googleapis.com, --allow-read and --allow-env alongside the env file. Docker is the alternative — docker compose run --rm deno-gemini-grounding-mcp-server as the client command, since MCP needs stdin and stdout. Build the image from a clone first — docker build -t deno-gemini-grounding-mcp-server . — since no published image exists.
Setup effort
Run a container — build the image from the repository, then point your client at it
USE CASES
01
Answering a question that needs current sources
Ask through the grounding tool so the answer is backed by a search rather than recall.
02
Keeping an API-calling server on a short leash
Run it under Deno with the network permission pinned to the one host it needs.
03
Running it from a client that only speaks stdio
Use the docker compose run --rm form so each call gets a fresh container over stdin and stdout.
04
Checking it works before wiring a client
Run the bundled Deno test tasks and watch the response validation output.
3 tools
Listed in project docs. Your client may show a different list.
call_gemini_or_vertex_ai
Calls either the Gemini Developer API or Vertex AI with a user prompt and options.
get_random_greeting
Returns a random greeting
BEFORE YOU INSTALL
What to know first
Tools you get
3 tools named in the project's documentation.
Requires
- docker on PATH
- GEMINI_API_KEY in your environment
Where it runs
The documented client runs the server on your machine.
INSTALLATION
Connect your client
Build the image from the repository, then point your client at it.
Server setup
TERMINAL
git clone https://github.com/rinerebox1/deno-gemini-grounding-mcp-server
cd deno-gemini-grounding-mcp-server
docker build -t deno-gemini-grounding-mcp-server .Then point your client at the image you just built — the block under MCP Clients runs it with your GEMINI_API_KEY
MCP Clients
Any MCP client — Docker
ANY MCP CLIENT — DOCKER · JSON
{
"mcpServers": {
"deno-gemini-grounding-mcp-server": {
"command": "docker",
"args": [
"run",
"-i",
"--rm",
"-e",
"GEMINI_API_KEY=<YOUR_GEMINI_API_KEY>",
"deno-gemini-grounding-mcp-server"
]
}
}
}<YOUR_GEMINI_API_KEY> is your Gemini API key. -i keeps stdin open, which a stdio MCP server needs; without it the container exits as soon as it starts.
