SERPdive MCP is a remote web search MCP server: add one URL, get one tool (serpdive_search) that returns clean page content sized for LLM context windows. 1,000 free searches per month, no credit card, no npm install required. On a public 1,000-question benchmark it wins 60.7% of decided blind quality duels against Tavily's default search while returning 20.2% fewer tokens, at 1–2 s per search.
Connect it in 30 seconds
Claude Code — one command:
claude mcp add --transport http serpdive https://mcp.serpdive.com \
--header "Authorization: Bearer sd_live_YOUR_KEY"Claude Desktop, Cursor, Windsurf, VS Code — any client that takes a JSON config or a server URL:
{
"mcpServers": {
"serpdive": {
"url": "https://mcp.serpdive.com/?key=sd_live_YOUR_KEY"
}
}
}Older clients that only speak stdio (no remote-server support) can run the same tool locally instead: npx serpdive-mcp — a zero-dependency npm package. The hosted URL stays the recommended path; the package exists so no client is left out.
Auth is your regular API key — as an Authorization: Bearer header, or as a ?key= parameter for clients that only accept a URL. Keys are created free in the dashboard, and each key can carry its own expiration date and monthly credit cap, which matters when you ship a key inside a tool you don't fully control.
Why a hosted server beats an npx wrapper
Most web search MCP servers are npm packages: npx something-mcp, an environment variable for the key, a Node process per client, per machine, per teammate. That is all accidental complexity — the package is just a proxy to a cloud API. A hosted server removes the whole layer:
- Zero install — no Node version conflicts, no package updates, no background processes.
- One config for the whole team — the same URL works on every machine and in CI.
- Server-side upgrades — extraction improvements land without you touching anything.
- Key hygiene — per-key spend caps and instant revocation from the dashboard, instead of a raw key in every teammate's env file.
What comes back matters more than the transport
Most search tools hand your agent a list of URLs and snippets, and the agent burns turns (and tokens) fetching and cleaning pages. SERPdive does the reading server-side: the tool returns the sentences that answer, already extracted from live pages, sized for a context window. That difference is measured, not asserted:
SERPdive MCP vs Tavily, Exa, Firecrawl and Brave MCP
All competitor details as of July 2026, from public docs and pricing pages.
| SERPdive MCP | Tavily MCP | Exa MCP | Firecrawl MCP | Brave Search MCP | |
|---|---|---|---|---|---|
| Hosted (remote) server | Yes — plus npx for stdio-only clients | Yes + npx option | Yes + npx option | Yes + npx option | npx (local) primarily |
| Results are | Extracted, answer-ready content | Snippets (raw content optional) | Semantic results + contents | Scraped pages (scrape-first tools) | Links + snippets |
| Free tier | 1,000 credits/mo, no card | 1,000 credits/mo | Trial credits | 1,000 credits/mo (scraping credits) | $5/mo credit, card required |
| Search price after free | $5 / 1k credits (launch) | $8 / 1k basic, $16 / 1k advanced (payg) | $7 / 1k + $1 / 1k pages for contents | 2 credits / 10 results + scrape cost | $5 / 1k requests |
| Measured content quality | 60.7% decided duels vs Tavily basic (public repo) | — | — | — | — |
- You need site scraping and crawling tools in the same server (whole-site ingestion, JS rendering): Firecrawl MCP is built for exactly that.
- You want semantic / similarity search — “pages like this one”, people and company lookup: Exa MCP exposes Exa's neural index.
- You have a hard requirement for an independent index with a privacy posture: Brave's server queries Brave's own index.
- You want no account at all: open-source multi-engine servers exist on GitHub — you trade setup time and result quality for $0.
1,000 free searches every month, no credit card. Hosted MCP server — paste the URL, get answer-ready web content.
Get your free API keyFrequently asked questions
What is a web search MCP server?
MCP (Model Context Protocol) is the open standard AI assistants use to call external tools. A web search MCP server exposes a search tool over that protocol, so clients like Claude Code, Claude Desktop, Cursor or Windsurf can query the live web mid-conversation. The quality of what comes back — links versus extracted content, and how many tokens it costs — depends entirely on the search engine behind the server.
Do I need an API key?
Yes — a free SERPdive key, created in the dashboard in about a minute. Every account gets 1,000 free credits per month with no credit card, and one Mako search costs 1 credit, so the free tier is 1,000 real searches a month.
Is there a free web search MCP server?
SERPdive's hosted MCP server is free within the monthly 1,000-credit allowance — no card required. Among competitors (as of July 2026): Tavily also gives 1,000 free credits/month, Exa gives trial credits, and Brave offers $5 of monthly credit but requires a credit card even on the free plan.
Which web search MCP server works with Claude Code?
Any server speaking the protocol works; the difference is setup. SERPdive is a single claude mcp add command because it is hosted over HTTP — no npm install, no local process. Most alternatives ship as npx packages you run locally, though several now also offer hosted endpoints.
Can I use it with Cursor, Windsurf or VS Code?
Yes. Every MCP client that accepts a server URL works with the JSON config above — put your key in the ?key= parameter if the client cannot send custom headers. That covers Cursor, Windsurf, VS Code with MCP-enabled extensions, and Claude Desktop.
What is the difference between a remote (hosted) and a local MCP server?
A local server is a process you install and keep running per machine — an npm package, its updates, its environment variables. A remote server is just a URL: the provider runs it, every machine and teammate uses the same line of config, and nothing breaks when Node versions change. For a tool that calls a cloud API anyway, running a local wrapper adds moving parts without adding capability.
Does the MCP server cost the same as the API?
Exactly the same: a search through MCP is billed like the same search through POST /v1/search — same credits (Mako 1, Moby 1.5), same rate limits, same dashboard usage view. There is no MCP surcharge.
Does the choice of search MCP server really affect agent output quality?
Measurably. The context your agent reads is the ceiling of what it can answer. In blind duels on a public 1,000-question benchmark, SERPdive's results won 60.7% of decided rounds against Tavily's default search while using 20.2% fewer tokens — same questions, same judge, methodology in the open repo. Better input, cheaper context window, better answers.