SERPdive Web Search — Claude Code MCP server: how to install
Connects web search to MCP clients with extracted, clean page content sized for LLMs. Useful when your agent needs current facts without drowning context in raw markup.
claude mcp add --transport http serpdive-mcp https://mcp.serpdive.com/mcpWhat SERPdive Web Search does
The server exposes a single tool called serpdive_search. It handles incoming search queries in any language with automatic localization. Instead of dumping raw HTML, the backend extracts the relevant text from top pages and returns a clean JSON payload containing titles, links, timestamps, and extracted content. The engine deliberately skips generating a final answer, letting your model synthesize results directly without an extra network round-trip.
Search requests run against two models. By default, mako extracts the primary factual sentences from each result, consuming 1 credit. The moby alternative costs 1.5 credits and returns entire page texts, which takes about one additional second. For general question answering, moby bloats the context window with unnecessary tokens.
You can connect via the hosted endpoint at https://mcp.serpdive.com/mcp or run locally with npx -y serpdive-mcp. Local execution uses Node.js and requires a SERPDIVE_API_KEY, which is available without entering payment card details.
What it helps with
- retrieve up-to-date web information directly from agent prompts
- extract clean article content without clutter or raw HTML tags
- reduce token consumption when pulling external reference materials
- provide verified factual excerpts for model summarization
Example prompts for the agent
- Run
serpdive_searchto find recent changes in the Node.js 22 runtime. - Check recent developments in quantum computing using the
makomodel. - Retrieve full article text for deep research on distributed databases with
moby.
How to install SERPdive Web Search
1Platform CLI: remote server
claude mcp add --transport http serpdive-mcp https://mcp.serpdive.com/mcp2Platform CLI
claude mcp add --env SERPDIVE_API_KEY=<SERPDIVE_API_KEY> --transport stdio serpdive-web-search -- npx -y serpdive-mcpReplace <…> with your own values
1Platform CLI: remote server
codex mcp add serpdive-web-search --url https://mcp.serpdive.com/mcp2Platform CLI
codex mcp add serpdive-web-search --env SERPDIVE_API_KEY=<SERPDIVE_API_KEY> -- npx -y serpdive-mcpReplace <…> with your own values
Config file
{
"mcpServers": {
"serpdive-web-search": {
"url": "https://mcp.serpdive.com/mcp"
}
}
}{
"mcpServers": {
"serpdive-web-search": {
"command": "npx",
"args": [
"-y",
"serpdive-mcp"
],
"env": {
"SERPDIVE_API_KEY": "<SERPDIVE_API_KEY>"
}
}
}
}Replace <…> with your own values
Config file
{
"servers": {
"serpdive-web-search": {
"type": "http",
"url": "https://mcp.serpdive.com/mcp"
}
}
}{
"servers": {
"serpdive-web-search": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"serpdive-mcp"
],
"env": {
"SERPDIVE_API_KEY": "<SERPDIVE_API_KEY>"
}
}
}
}Replace <…> with your own values
1Platform CLI: remote server
gemini mcp add -t http serpdive-web-search https://mcp.serpdive.com/mcp2Platform CLI
gemini mcp add -e SERPDIVE_API_KEY=<SERPDIVE_API_KEY> serpdive-web-search npx -y serpdive-mcpReplace <…> with your own values
Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about SERPdive Web Search
How do I install SERPdive Web Search in Claude Code?
Run claude mcp add --transport http serpdive-web-search https://mcp.serpdive.com/mcp in your terminal to connect to the hosted HTTP endpoint.
What does the serpdive_search tool return?
It returns structured JSON containing search results with URLs, page titles, dates, and extracted textual content. It does not generate synthetic answers on the server side, leaving reasoning to your client LLM.
What is the difference between mako and moby models?
The default mako model costs 1 credit and extracts only concise, fact-bearing sentences. The moby model costs 1.5 credits and returns the full page content, which adds around 1 second of latency and uses significantly more tokens.
