deep_search
Perform a deep search on a given query using an advanced web research agent.
LABSCO SUMMARY
One call in place of a search-read-search loop, at the cost of not steering it.
The entire surface is a query and an effort level, so this is a tool you delegate to rather than drive: which pages get read is the agent's decision, not yours. That fits a broad question you want answered well, and does not fit a narrow one where you already know which source should be consulted.
What it is
A single-tool server that runs a web research agent over a query, with an effort setting that trades depth against time.
What you get
- A question researched by an agent that decides what to search and what to read, with the answer and its sources written to a JSON file whose path comes back — over stdio the server returns the path rather than the text, to save tokens
- An effort argument, so a quick check and a thorough pass are the same call with a different setting
Requirements
GEMINI_API_KEY. The research runs against Google's Gemini API under your own key, and there is nothing else to configure — the server carries one tool.
Setup effort
Add the configuration to your client
USE CASES
01
Answering a question the model's own knowledge does not cover
Hand over the question as it stands and let the agent do the searching and reading, for topics where you would otherwise open six tabs to build the same answer.
02
Deciding how much research a question deserves
Set effort low when you want a quick orientation and high when the answer is going into something that matters — the same call, with the depth as your decision rather than the agent's.
1 tool
Counted once with only startup setup. Other versions, modes or optional credentials may expose a different list.
BEFORE YOU INSTALL
What to know first
Tools you get
1 tool when we connected. The effort argument sets cost: low runs 1 query and 1 loop on Flash, medium 3 queries and 2 loops on Flash, high 5 queries and 3 loops on Pro. Over stdio it returns a file_path, not the answer.
Requires
- uvx on PATH
- Python 3.12+
- GEMINI_API_KEY in your environment
Where it runs
The documented client runs the server on your machine.
CONTEXT COST
Light Standing cost
Rated by the size of the one tool definition that rides along on every turn
INSTALLATION
Connect your client
MCP Clients
Claude Desktop
CLAUDE DESKTOP · JSON
{
"mcpServers": {
"gemini-deepsearch": {
"command": "uvx",
"args": [
"gemini-deepsearch-mcp"
],
"env": {
"GEMINI_API_KEY": "<YOUR_GEMINI_API_KEY>"
},
"timeout": 180000
}
}
}
