Quick answer: link.sc runs a hosted Model Context Protocol server at https://mcp.link.sc/. Add that URL and your link.sc API key to Claude Desktop or Claude Code, and Claude gains two tools: search (Google results as markdown) and fetch (any URL as clean markdown). There is nothing to install and nothing to run locally.
What "hosted MCP server" means
Most MCP servers you see are local: you install a package, Claude launches it as a subprocess, and it talks over stdio. That works, but it means every teammate installs and updates the same binary, and the server runs on their machine.
A hosted MCP server is different. It runs as a remote HTTP endpoint that any MCP client connects to over the network. You add a URL and a key, and you are done. link.sc exposes exactly this at https://mcp.link.sc/, backed by the same data engine and API key as the REST API.
Get your API key
Sign in at link.sc/dashboard/keys and create a key. It starts with lsc_. Keep it somewhere safe, and treat it like a password: anyone with the key can spend your quota.
Set it up in Claude Desktop
Open your claude_desktop_config.json and add the server, then restart Claude:
{
"mcpServers": {
"linksc": {
"url": "https://mcp.link.sc/",
"transport": "http",
"headers": {
"x-api-key": "lsc_your_api_key"
}
}
}
}
The config file lives here:
- macOS:
~/Library/Application Support/Claude/claude_desktop_config.json - Windows:
%APPDATA%\Claude\claude_desktop_config.json
Set it up in Claude Code
One command (add -s user to make it available in every project, not just this one):
claude mcp add --transport http linksc https://mcp.link.sc/ \
--header "x-api-key: lsc_your_api_key"
Check that it connected:
claude mcp list
# linksc ✓ connected
Set it up in claude.ai (web)
On a Pro, Max, Team, or Enterprise plan, open Settings, then Connectors, then Add custom connector, name it linksc, and paste the URL https://mcp.link.sc/. One caveat: the claude.ai connector screen is built around OAuth and has no field for a custom header, so the x-api-key approach shown above is easiest in Claude Desktop or Claude Code today. Use those two if you want key-based auth right now.
Check that it works
Ask Claude something that needs the live web and name the tool so it reaches for it:
"Using linksc, search for the latest Claude model releases and give me the top three with links."
Claude runs search, reads the results, and answers from them. Point it at a page the same way: "Using linksc, fetch this URL and summarize it."
The two tools
Once connected, Claude can call:
- search: give it a query and it returns Google results as markdown, including organic results, People Also Ask questions, and shopping results when they appear.
- fetch: give it a URL and it returns the page as clean markdown. It renders JavaScript, gets past most anti-bot walls, and automatically parses Google and Bing result pages into structured data.
You do not call these yourself. You ask Claude a question, and it decides when to reach for the web:
"Search for the latest pricing on the top three managed Postgres providers and make a comparison table."
Claude runs search, reads the results, fetches the pricing pages it needs, and answers from what it actually found instead of from its training data.
Why route web access through link.sc
The hard part of web access is not the HTTP request. It is getting a real page back. Sites throw JavaScript challenges, bot detection, and rate limits at raw requests. link.sc handles the fingerprinting, residential egress, and browser rendering behind one endpoint, so Claude sees the same clean markdown whether the target is a static blog or a Cloudflare-protected app.
You also get one bill and one quota across the REST API and the MCP server, rather than juggling separate scraping infrastructure.
Try it without Claude
The server speaks JSON-RPC 2.0 over the Streamable HTTP transport, so you can poke it with curl. Listing tools needs no key:
curl -X POST https://mcp.link.sc/ \
-H "Content-Type: application/json" \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'
Calling a tool needs your key:
curl -X POST https://mcp.link.sc/ \
-H "Content-Type: application/json" \
-H "x-api-key: lsc_your_api_key" \
-d '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"search","arguments":{"q":"anthropic claude"}}}'
Billing
MCP tool calls cost the same as REST API requests: one request per fetch or search. You can watch usage in real time at link.sc/dashboard/usage.
Where to go next
New to MCP itself? Start with What Is the Model Context Protocol (MCP) and Why It Matters. Building an agent instead of using Claude directly? See How to Give an AI Agent Access to the Internet. Full reference lives in the API docs.