search-online-mcp
Exposes a standardized web search tool via Brave Search for MCP clients like Codex, Claude, and Cursor, returning stable JSON.
README
[!NOTE] Vibe-coded with Codex. Use it at your own risk.
Search Online MCP Server
An MCP (Model Context Protocol) server that exposes a standardized web search tool for agents and clients like OpenAI Codex, Claude, and Cursor. It uses Brave Search under the hood and returns stable, agent‑friendly JSON.
Features
- One tool:
search_online(engine: Brave; verticals: web/news/images/videos) - Stable output shape via
structuredContentand a concisetextsummary - Fallback chain for robustness:
- Fish function
search_online(if available) uvx brave-search-python-client(httpx/psutil)- Direct Brave HTTP API (X-Subscription-Token)
- Fish function
- No stdout leakage: safe JSON‑RPC framing (all debug goes to logs)
Install
cd ~/Projects/mcp
git clone <repo> search-online-mcp
cd search-online-mcp
npm install
Configure Codex
~/.codex/config.toml:
[mcp_servers.searchonline]
command = "node"
args = ["/Users/YOU/Projects/mcp/search-online-mcp/index.mjs"]
env = {
BRAVE_SEARCH_PYTHON_CLIENT_API_KEY = "YOUR_BRAVE_KEY",
# skip fish/jq, go straight to uvx for stability
SEARCH_ONLINE_FORCE_UVX = "1",
}
startup_timeout_ms = 20_000
tool_timeout_sec = 60
Restart Codex. Verify:
codex mcp list
codex mcp get searchonline --json
Optional: install the Fish function directly
If you use fish shell, you can install the search_online function for your own terminal use. It’s the same logic the MCP server’s “fast path” expects.
Dependencies: fish 3+, uvx (from astral-sh/uv), jq, glow (optional), and a Brave API key.
Install:
mkdir -p ~/.config/fish/functions
cp contrib/fish/search_online.fish ~/.config/fish/functions/search_online.fish
set -Ux BRAVE_SEARCH_PYTHON_CLIENT_API_KEY "YOUR_BRAVE_KEY"
# Open a new shell or: source ~/.config/fish/functions/search_online.fish
Usage examples:
# JSON (default)
search_online -L 5 "gaston morixe" | jq '.results[:3] | map({title,url})'
# NDJSON streaming
search_online -o ndjson -L 3 "openai codex" | head -n 3
# Plain text (no color)
search_online -o text -n "rust async tutorial"
# Raw engine JSON
search_online -o raw "openai codex" | jq '.web.results[:2] | map({title,url})'
Notes:
- The function prints a standardized JSON object in
-o jsonmode (engine/vertical/query/results…). - Set
SEARCH_ONLINE_ENGINEto override default engine (currently onlybrave). glowis optional;-n/--no-colordisables colorized output.
Tool schema
Input (zod → JSON Schema):
querystring (required) — search queryverticalenum: web|news|images|videos (default web)limitinteger 1..20 (optional)offsetinteger ≥0 (optional)country,lang,market(optional)
Output:
content:[ { type: "text", text: "ok results=N" } ]structuredContent: full JSON with keys:engine,vertical,query,fetched_at,results[](rank, title, url, snippet, site_name, etc.)- on error:
isError: truewith a text explanation
Environment variables
BRAVE_SEARCH_PYTHON_CLIENT_API_KEY— Brave API key (required)SEARCH_ONLINE_FORCE_UVX=1— skip fish; go straight touvxSEARCH_ONLINE_DEBUG_STDERR=1— mirror debug to stderr (Codex logs)
Logs
- MCP server writes to
~/.codex/log/search_online_mcp.log - Codex session logs:
- TUI:
~/.codex/log/codex-tui.log - exec (non-interactive): stderr you redirect (e.g.,
--json >out 2>err)
- TUI:
Tests
Local tests (require Brave key unless you mock):
npm run test:all
What they do:
tests/list_names.mjs: confirms tool names comply with the allowed patterntests/call_basic.mjs: callssearch_online(usesstructuredContent)tests/call_invalid.mjs: confirms missingqueryproduces an MCP error
Troubleshooting
- See
~/.codex/log/search_online_mcp.logfor per-call details (PATH, cmd, uvx/http status). - If you still see “tool call failed” in Codex:
- Kill stale server:
pkill -f "search-online-mcp/index.mjs" - Start a new Codex session (not resume)
- Ensure
SEARCH_ONLINE_FORCE_UVX=1
- Kill stale server:
License
MIT
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.