Serper.live

Serper.live MCP server

Live Google search for AI agents. This Model Context Protocol server lets Claude, OpenAI Codex, Cursor, VS Code and other MCP clients run a real Google search through Serper.live for the country, language and device you choose, and check where a domain ranks right now.

Every search is fresh: nothing is cached, and a page of 10 results usually comes back in 2–3 seconds. Positions count organic results only, without ads, sitelinks, video or image blocks, so the number an agent reads is the number a rank tracker would store.

Tools

Arguments: query (2–120 characters), country (two-letter code such as US, GB, DE, FR, TR, UA), optional language (Google interface language, defaults to the country's), device (desktop or mobile) and limit / depth (1–100, default 10).

Setup

1. Create an account at serper.live: it starts with 100 free searches, no card needed. Then create an API token on API tokens. Add credit when the free searches run out: by card from $10 or in USDT or USDC from $5, no subscription, credit never expires.
2. Add the server to your client with the token in SERPER_LIVE_TOKEN. It runs with npx, so Node.js 20 or newer is all you need; Claude Desktop can also install it in one click without Node.js.

Claude Code

claude mcp add serper-live -e SERPER_LIVE_TOKEN=sl_live_YOUR_TOKEN -- npx -y @serper-live/mcp

Claude Desktop

One click: download serper-live-mcp.mcpb from the latest release and open it. Claude Desktop shows an install dialog, asks for the API token once, stores it securely and runs the server with its own Node.js.

Or by hand, in claude_desktop_config.json:

{

"mcpServers": { "serper-live": { "command": "npx", "args": ["-y", "@serper-live/mcp"], "env": { "SERPER_LIVE_TOKEN": "sl_live_YOUR_TOKEN" } } } }

Cursor

~/.cursor/mcp.json or .cursor/mcp.json in a project:

{

"mcpServers": { "serper-live": { "command": "npx", "args": ["-y", "@serper-live/mcp"], "env": { "SERPER_LIVE_TOKEN": "sl_live_YOUR_TOKEN" } } } }

VS Code

.vscode/mcp.json — VS Code asks for the token once and keeps it out of the file:

{

"inputs": [ { "type": "promptString", "id": "serper-live-token", "description": "Serper.live API token", "password": true } ], "servers": { "serper-live": { "type": "stdio", "command": "npx", "args": ["-y", "@serper-live/mcp"], "env": { "SERPER_LIVE_TOKEN": "${input:serper-live-token}" } } } }

OpenAI Codex

codex mcp add serper-live --env SERPER_LIVE_TOKEN=sl_live_YOUR_TOKEN -- npx -y @serper-live/mcp

Or by hand, in ~/.codex/config.toml (the Codex CLI, its IDE extension and the ChatGPT desktop app share this file):

[mcp_servers.serper-live]

command = "npx" args = ["-y", "@serper-live/mcp"]

[mcp_servers.serper-live.env] SERPER_LIVE_TOKEN = "sl_live_YOUR_TOKEN"

Any other MCP client: run npx -y @serper-live/mcp over stdio with SERPER_LIVE_TOKEN in the environment. Agents that set up servers from a repository, such as Cline, follow llms-install.md.

Environment

What to ask

• "Where does example.com rank on Google in Germany on mobile for *project management software*? Check the top 30."
• "Compare the top 10 Google results for *best running shoes* in the US and the UK."
• "Which markets can you search in?"

Billing and retries

Searches are billed to your Serper.live balance per page of up to 10 organic results that received results, at the price of the market (pricing); desktop and mobile cost the same. The free searches of the account are used first, a search that finds nothing costs nothing, and a search that fails or runs out of time is never charged. Each call sends an Idempotency-Key, so a repeated call is never billed twice.

Development

npm install

npm run build npx @modelcontextprotocol/inspector node dist/index.js npm run bundle # serper-live-mcp.mcpb: dist/, manifest.json, icon.png and the production dependencies

Releasing

The version lives in package.json, in version plus packages[0].version of server.json, which the MCP Registry checks against npm, and in manifest.json of the MCP Bundle. npm version sets all of them: its version script (scripts/sync-version.mjs) copies the new version into server.json and manifest.json.

A pushed v* tag runs .github/workflows/publish.yml: it checks that the tag matches every version field, builds, packs and validates the MCP Bundle, publishes to npm through trusted publishing (no npm token, with provenance), waits until npm serves the new version (up to 30 minutes), attaches serper-live-mcp.mcpb to the GitHub release and updates the MCP Registry through GitHub OIDC with two packages: npm and the bundle with its SHA-256. A failed run can be re-run as is: a version npm already has is not published again, and the release asset is replaced.

npm version x.y.z --no-git-tag-version   # package.json, package-lock.json, server.json, manifest.json

git commit -am "Release x.y.z" git tag vx.y.z git push && git push origin vx.y.z

The npm package lists this workflow as its trusted publisher (package settings → Trusted Publisher → GitHub Actions: serper-live/serper-live-mcp, workflow publish.yml, with Allow npm publish ticked under Allowed actions).

Links

• Serper.live — live Google SERP API
• API documentation
• Playground — try a live search in the browser without an account
• Contact

License

MIT