o3-search
An AI agent with advanced web search capabilities. Useful for finding the latest information, troubleshooting errors, and discussing ideas or design challenges. Supports natural language queries.
LABSCO SUMMARY
The searching happens behind the key instead of inside your transcript.
One tool, one input string: the fetching and reading that a question needs happen on the other side of the API key rather than as a chain of page loads filling your context. That is also the whole configuration surface — no result count, no site filter, no freshness window — so the question you write is the only lever you have.
What it is
A single-tool bridge to a web-searching model, called with a plain-language question.
What you get
- A natural-language question answered by a model that searches the web, in one call
- Current information for topics your own model has no training data for
- Error messages and stack traces researched without leaving the session
- A design question or half-formed idea talked through in the same call shape as a lookup
Requirements
An OpenAI API key in OPENAI_API_KEY, and a Node runtime.
Setup effort
One command plus a key — npx o3-search-mcp, then supply credentials
USE CASES
01
Getting to the bottom of an error you have not seen before
Hand over the message and what you were doing when it appeared. The searching and reading happen behind the call, so what returns is an answer rather than a list of pages to open.
02
Checking what changed in a library since your model last knew
Ask in plain language for the current state of something your own model can only guess at, and the answer comes from a search made now rather than from training data.
03
Talking through a design decision before committing to it
The tool takes an open question as readily as a lookup, so a half-formed approach can be argued with rather than only fact-checked.
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 tool is o3-search, and OPENAI_MODEL switches it between o3 (the default), o4-mini and gpt-5. Calling o3 needs an OpenAI key on tier 4 or a verified organization; otherwise every call errors.
Requires
- npx on PATH
- OPENAI_API_KEY in your environment
CONTEXT COST
Light Standing cost
Rated by the size of the one tool definition that rides along on every turn
INSTALLATION
Connect your client
Server setup
TERMINAL
npx o3-search-mcpMCP Clients
Claude Code — npx
CLAUDE CODE — NPX · TERMINAL
claude mcp add o3 -s user -e OPENAI_MODEL=o3 -e OPENAI_API_KEY=<YOUR_OPENAI_API_KEY> -e SEARCH_CONTEXT_SIZE=medium -e REASONING_EFFORT=medium -e OPENAI_API_TIMEOUT=300000 -e OPENAI_MAX_RETRIES=3 -- npx o3-search-mcpAny MCP client — npx
ANY MCP CLIENT — NPX · JSON
{
"mcpServers": {
"o3-search": {
"command": "npx",
"args": [
"o3-search-mcp"
],
"env": {
"OPENAI_API_KEY": "<YOUR_OPENAI_API_KEY>",
"OPENAI_MODEL": "o3",
"SEARCH_CONTEXT_SIZE": "medium",
"REASONING_EFFORT": "medium",
"OPENAI_API_TIMEOUT": "300000",
"OPENAI_MAX_RETRIES": "3"
}
}
}
}Any MCP client — local build
ANY MCP CLIENT — LOCAL BUILD · TERMINAL
git clone https://github.com/yoshiko-pg/o3-search-mcp.git
cd o3-search-mcp
pnpm install
pnpm buildPoint your client at build/index.js inside that folder:
ANY MCP CLIENT — LOCAL BUILD · JSON
{
"mcpServers": {
"o3-search": {
"command": "node",
"args": [
"<YOUR_PATH_TO_O3_SEARCH_MCP>/build/index.js"
],
"env": {
"OPENAI_API_KEY": "<YOUR_OPENAI_API_KEY>",
"OPENAI_MODEL": "o3",
"SEARCH_CONTEXT_SIZE": "medium",
"REASONING_EFFORT": "medium",
"OPENAI_API_TIMEOUT": "300000",
"OPENAI_MAX_RETRIES": "3"
}
}
}
}
