Labsco
mixelpixx logo

Google Research

โ˜… 248

from mixelpixx

Perform advanced web research using Google Search, with intelligent content extraction and multi-source synthesis.

๐Ÿ”ฅ๐Ÿ”ฅ๐Ÿ”ฅ๐Ÿ”ฅโœ“ VerifiedAccount requiredAdvanced setup

Google-Research-MCP

A standalone Windows MCP server that gives Claude Google search and web research capabilities, powered by SerpAPI. It ships as a single self-contained .exe โ€” no Node.js, no npm install, no external runtime.

Tools

ToolDescription
google_searchSearch Google and get quality-scored results with authority ratings
extract_webpage_contentExtract clean readable content from any URL
extract_multiple_webpagesExtract up to 5 URLs concurrently
research_topicMulti-source research that returns a structured synthesis prompt

The binary also bundles six research skills and one research agent that Claude can invoke automatically. These are installed into your ~/.claude directory by init (see Skills and agents).

Getting a SerpAPI key

  1. Go to serpapi.com and create a free account.
  2. Copy your API key from the dashboard.
  3. The free plan includes 100 searches/month.

CLI subcommands

Run these from cmd or PowerShell:

CommandDescription
(none)Starts the MCP server when launched by a client; opens the GUI installer when double-clicked from Explorer; prints help when run in a terminal
initInstall the skills and agent into ~/.claude/skills and ~/.claude/agents
uninstallRemove the installed skills and agent
statusShow install status
diagnoseSelf-test: makes one live SerpAPI request and reports timing
service installSave the key, register HTTP autostart, and start the background server
service uninstallRemove the autostart registration
service statusShow config, autostart, and port state
service startSpawn the background HTTP server now (no autostart)
skill <name>Print a skill's markdown to stdout
--versionShow version
--helpShow usage

The server also re-installs the bundled skills on first launch as a safety net, so they stay current even if you forget to re-run init after updating the exe.

GUI installer

Double-clicking google-research-mcp.exe from File Explorer opens a native install dialog. Paste your SerpAPI key, optionally enable the background HTTP service, and click Install. This writes the skills and agent, saves your key to the config file, and (if enabled) registers the HTTP service to start on login. The background service runs as a separate windows-subsystem binary, google-research-mcp-tray.exe, which must sit next to the main exe.


Environment variables

VariableRequiredDefaultDescription
SERPAPI_KEYYes*โ€”Your SerpAPI key
MCP_TRANSPORTNostdioTransport mode: stdio or http
PORTNo3030HTTP server port (used only when MCP_TRANSPORT=http)

* If SERPAPI_KEY is not set, the server falls back to the key stored in the config file (see below). One of the two must be present.

Config file

When you install the HTTP service (via the GUI or service install), your key and port are saved to:

Copy & paste โ€” that's it
%USERPROFILE%\.google-research-mcp\config.toml

This file is used when the server is launched by Windows autostart, where no env block is available. The key is stored in plaintext, so if that is a concern, prefer passing the key through your client's env block and skip the HTTP service. uninstall does not delete this file โ€” remove it manually if you want to wipe the saved key.


HTTP mode

For HTTP-based MCP clients (for example LM Studio), run the server with the HTTP transport:

Copy & paste โ€” that's it
$env:SERPAPI_KEY = "your_key"
$env:MCP_TRANSPORT = "http"
$env:PORT = "3030"
.\google-research-mcp.exe

The endpoint is then available at http://localhost:3030/mcp:

Copy & paste โ€” that's it
{
  "mcpServers": {
    "google-research": {
      "url": "http://localhost:3030/mcp"
    }
  }
}

For an always-on background service with autostart, use service install instead of running the exe manually. See binary-release/HTTP_SETUP.md for the full walkthrough.

Security note

The HTTP transport binds to 0.0.0.0 (all network interfaces) and the /mcp endpoint is not authenticated. Anyone who can reach the port can call the tools and consume your SerpAPI quota. Only enable HTTP mode on a trusted network or behind a firewall that restricts the port to localhost. The default stdio mode (used by Claude Desktop and Claude Code) is not network-exposed and is unaffected.


Skills and agents

init (and the GUI installer) write these into your ~/.claude directory. Claude invokes them automatically based on your intent; you can also call a skill explicitly with /skill-name.

Skills (~/.claude/skills):

SkillPurpose
deep-researchMulti-source research with structured synthesis
fact-checkVerify a claim against multiple authoritative sources
find-docsLocate official documentation and API references
news-monitorTrack recent news and developments on a topic
competitive-analysisResearch competitors and market landscape
cite-sourcesFormat citations and references

Agent (~/.claude/agents):

AgentPurpose
deep-research-agentRuns a heavy research task in an isolated context and returns only a summary, keeping the main conversation clean

Tool reference

google_search

Search Google and return ranked, quality-scored results.

ParameterTypeRequiredDescription
querystringYesThe search query
num_resultsnumberNoNumber of results (default 5, max 10)
sitestringNoRestrict to a domain, e.g. wikipedia.org
languagestringNoISO 639-1 code, e.g. en, fr, de
date_restrictstringNod7 (7 days), w2 (2 weeks), m6 (6 months), y1 (1 year)
exact_termsstringNoPhrase that must appear in results
result_typestringNonews, image, or video (default: web)
pagenumberNoPage number, 1-based (default 1)
results_per_pagenumberNoResults per page (default 5, max 10)
sortstringNorelevance (default) or date

extract_webpage_content

Fetch a URL and extract clean readable content, stripping navigation, ads, and boilerplate.

ParameterTypeRequiredDescription
urlstringYesFull URL (must start with http:// or https://)
formatstringNomarkdown (default), html, or text
full_contentbooleanNoReturn full content (true) or a 500-character preview (false, default)
max_lengthnumberNoMaximum character length to return

Returns: title, description, word count, summary, and content in the requested format.

extract_multiple_webpages

Extract content from up to 5 URLs concurrently.

ParameterTypeRequiredDescription
urlsstring[]YesArray of URLs (max 5)
formatstringNomarkdown (default), html, or text

Returns a summary (title, URL, word count, excerpt) for each URL.

research_topic

Searches for sources, extracts their content, and returns a structured synthesis prompt for Claude to analyze. No external AI API is called โ€” the synthesis is performed by Claude itself.

ParameterTypeRequiredDescription
topicstringYesThe topic to research
depthstringNobasic, intermediate (default), or advanced
num_sourcesnumberNoNumber of sources to consult (1โ€“10, default 5)
focus_areasstring[]NoSpecific subtopics, e.g. ["performance", "security"]

Depth guide:

  • basic โ€” 3 sources, short overview
  • intermediate โ€” 5 sources, key findings and themes
  • advanced โ€” 8โ€“10 sources, contradictions and recommendations

Building for distribution

Copy & paste โ€” that's it
cargo build --release

The resulting target\release\google-research-mcp.exe is fully self-contained โ€” no DLL dependencies beyond standard Windows system libraries. Copy it to any Windows 10/11 machine and run it directly. Build google-research-mcp-tray.exe alongside it if you need the background HTTP service.