Hosted MCP server

Web search MCP server for Claude, Cursor, and any MCP client

Point a client at one hosted endpoint and the model can search the web, read pages, follow the news, and see what is trending. Clients that speak OAuth 2.1 never need an API key in a config file.

Endpoint
https://mcp.search1api.com/mcp

What connecting gives you

The client reads the tool list when it connects, and the model calls one when it needs something from the live web.

Five read-only tools

search, news, crawl, sitemap and trending, advertised by the server itself.

No key in a config file

A client that speaks OAuth 2.1 approves in the browser and holds a token instead.

Nothing running locally

The server is hosted, so there is no process to install or keep alive.

Tools

search / news / crawl / sitemap / trending

What is a web search MCP server?

The Model Context Protocol is a standard for how an AI client describes and calls external tools. A web search MCP server implements that standard on top of a search backend: the client connects, reads the tool list, and the model calls a tool when it needs something from the live web. You do not write HTTP requests, thread a key through each call, or parse a result format. Search1API runs one as a hosted service, so there is no local process to install or keep alive.

Five tools your agent gets

The server advertises these when a client connects. Every one is a read-only call against the live web.

search

Search the web

Web search across 13 engines, including Google, Bing, DuckDuckGo, GitHub, Reddit, arXiv, YouTube, and WeChat. Each result carries an id, title, URL, and text the model can cite. Set crawl_results and the top matches come back as full page content in the same call.

news

Search current news

News search across Google, Bing, DuckDuckGo, Yahoo, and Hacker News, with an optional recency window of a day, a month, or a year.

crawl

Read a web page

Takes one public URL and returns the readable title and full text, without the navigation and boilerplate wrapped around it.

sitemap

Discover links on a site

Lists the public links found on a page or a domain, for when the agent needs to see how a site is laid out before reading any of it.

trending

Explore trending topics

Currently trending repositories on GitHub, or current stories on Hacker News.

Connect your client

Almost every MCP client takes the same JSON; what differs is the credential. Pick the shape that matches your setup.

No credential: OAuth 2.1

Give an OAuth-aware client the endpoint and nothing else. The server answers the first connection with a 401 pointing at its protected-resource metadata; the client discovers the authorization server, registers itself, and runs Authorization Code with PKCE. You approve it in a browser and no key is stored.

json
{
  "mcpServers": {
    "search1api": {
      "url": "https://mcp.search1api.com/mcp"
    }
  }
}

API key in a header

For a client that does not do OAuth discovery yet, create a key in the dashboard and send it as a bearer token. Same endpoint, same credits, same rate limits.

json
{
  "mcpServers": {
    "search1api": {
      "url": "https://mcp.search1api.com/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_SEARCH1API_KEY"
      }
    }
  }
}

Local stdio

The same server also runs on your machine straight from npm, so the MCP protocol never leaves it. The tool calls it makes still reach api.search1api.com.

json
{
  "mcpServers": {
    "search1api": {
      "command": "npx",
      "args": ["-y", "search1api-mcp"],
      "env": {
        "SEARCH1API_KEY": "YOUR_SEARCH1API_KEY"
      }
    }
  }
}

Where clients differ

Claude Code adds it in a single claude mcp add command rather than by editing a file.

VS Code uses servers rather than mcpServers, and requires type: "http" on the entry.

Windsurf and Claude.ai cannot send headers, so they take the key in the URL as ?apiKey=. That puts it in shell history and server logs, so prefer OAuth or a header wherever the client allows one.

Search1API is listed on Smithery, which offers a one-click install across 20+ clients.

How authentication works

The endpoint accepts an OAuth access token or a Search1API API key in the same Authorization: Bearer header. Both spend the same account credits at the same rates, so moving between them changes nothing about billing.

OAuth uses Authorization Code with PKCE (S256) and dynamic client registration, which means a client you have never set up can register itself on first contact. A client that asks for the offline_access scope can refresh its own token and stay connected across long sessions.

With no credential attached, the server returns 401 along with the metadata URL a compliant client needs to begin authorization. An expired token gets the same challenge with error="invalid_token", which is the signal to refresh rather than to fail.

A few clients can only put credentials in the URL, and the server accepts ?apiKey= for them. A key in a URL ends up in shell history and server logs, so use OAuth or a header wherever the client allows one.

What people use it for

A coding agent that checks current library docs and GitHub before writing against an API it has not seen.

A research session where the assistant searches first, then reads the three most useful results in full.

A morning brief assembled from news search with the recency window set to a day.

Competitive checks that read the competitor page directly instead of relying on what the model remembers.

Fact-checking a draft against sources the assistant can cite by URL.

FAQ

What is an MCP server for web search?

It is a server that implements the Model Context Protocol on top of a search backend. An MCP client connects, reads the list of tools the server offers, and the model calls one when it needs live information. The client handles transport and authentication once, so the model can search, read a page, or check the news without any per-call integration code.

Which clients work with it?

Any client that speaks the Model Context Protocol over Streamable HTTP. Configurations for Claude Code, Claude Desktop, Cursor, VS Code, Windsurf, and Claude.ai are on this page. Search1API is also listed on Smithery, which offers a one-click install across 20+ clients.

Do I need an API key?

Not if your client supports OAuth 2.1 discovery. Give it the endpoint, approve access in the browser, and the client holds a token instead of a key. For clients that do not do discovery yet, create a key in the dashboard and put it in an Authorization: Bearer header.

Which search engines can the search tool use?

Thirteen: google, bing, duckduckgo, yahoo, x, reddit, github, youtube, arxiv, wechat, bilibili, imdb, and wikipedia. The HTTP Search endpoint covers more than this, including Baidu, 360, and Quark; the MCP search tool exposes the thirteen listed here.

How much does it cost?

MCP tool calls are ordinary API calls and spend the same credits at the same rates. A search or news request is 1 credit, and each page successfully retrieved through crawl_results adds 1 more. New accounts get 100 credits free with no card, which is 100 searches.

Can I run it locally instead?

Yes. The same server runs over stdio straight from npm as search1api-mcp, using an API key. The MCP protocol then never leaves your machine, though the tool calls it makes still reach api.search1api.com.

Explore more

Give your agent live web access

One endpoint, five tools, 100 free credits to try it with. No credit card, and no local process to keep running.