You can give Claude Desktop live web search in a few minutes by pointing it at a hosted MCP server, no code, just a config block and a restart. Once connected, you can ask Claude to search Google, Reddit, YouTube, Amazon, or TikTok and it calls the tool itself instead of guessing from training data. This is the fastest way to ground a desktop assistant that keeps making things up about anything recent. The one detail people miss: the hosted MCP server authenticates with an x-api-key header, not the Bearer token the REST API uses. Get that right and the rest is copy-paste.
Prerequisites
- A Scavio API key (50 free credits to start)
- An MCP client: Claude Desktop, Cursor, Windsurf, or VS Code
- Two minutes to edit one config file
Walkthrough
Step 1: Add the hosted MCP server to your client config
Point your MCP client at the remote Scavio server and pass your key in the x-api-key header. This is the whole integration; there is no package to install for the hosted route.
{
"mcpServers": {
"scavio": {
"url": "https://mcp.scavio.dev/mcp",
"headers": { "x-api-key": "YOUR_SCAVIO_API_KEY" }
}
}
}Step 2: Restart the client and confirm the tools loaded
Fully quit and reopen the app so it reconnects. The Scavio search tools (Google, Reddit, YouTube, Amazon, Walmart, TikTok) should now appear in the client's tool or MCP panel.
# In Claude Desktop: quit fully, reopen, open the tools/MCP panel and look for scavioStep 3: Prompt it to search
Ask a question that needs current data. The client decides to call the tool, runs the search, and grounds its answer in the JSON that comes back instead of its training cutoff.
# Prompt: "Search Google for the current Firecrawl pricing tiers and summarize them."Step 4: Prefer a local server? Use the npm package
If you would rather run the server yourself instead of the hosted endpoint, install the published package and set your key as an environment variable. Same tools, local process.
npx -y @scavio/mcp-server # set SCAVIO_API_KEY in the client envPython Example
# MCP is configured in the client, not called from code. To hit the same data
# directly from Python, use the REST API with a Bearer token:
import os, requests
H = {"Authorization": f"Bearer {os.environ['SCAVIO_API_KEY']}"}
r = requests.post("https://api.scavio.dev/api/v1/google", headers=H, json={"query": "firecrawl pricing"})JavaScript Example
// The MCP config above is all most users need. For a code path, call REST directly:
const r = await fetch("https://api.scavio.dev/api/v1/google", {
method: "POST",
headers: { Authorization: `Bearer ${process.env.SCAVIO_API_KEY}`, "Content-Type": "application/json" },
body: JSON.stringify({ query: "firecrawl pricing" }),
});Expected Output
After the restart, the client lists the Scavio tools and can call them mid-conversation. A grounded answer cites live search results rather than the model's cutoff. Remember: MCP uses the x-api-key header, while the REST API uses Authorization: Bearer.