M
MCP Directory CURATED DIRECTORY
911 MCP servers indexed
中文
← Back to directory
G

Grok Search MCP

Community
Integrate Grok's powerful real-time search capabilities into Claude via MCP
How to install ↓ Compare GitHub source repository ↗
Category
Other
Stars
★ 1.9k Very popular
Transport
stdio (local process)
Runtime
Python 3.10+ · Prebuilt binary
Credentials
API key / credential required
License
MIT
Last commit
⚠
Tools
8
48FMRS · D

This project provides a dual-engine search (Grok + Tavily) with rich features and easy installation, suitable for Claude Code users. Strengths include intelligent search and automatic fallback mechanisms, but it relies on external APIs and requires key configuration.

Strongest · Documentation 12/20 Weakest · Reliability 8/20

Reliability
8/20
Security and permissions
8/20
Maintenance
10/20
Documentation
12/20
Setup experience
10/20
Why each score
Reliability 8/20
Source code shows a FastMCP-based server with tools matching README descriptions, but no tests or CI configuration to verify critical paths, limiting reliability to 8.
Security and permissions 8/20
API keys are passed via environment variables, README shows masked examples, no hardcoded secrets found. However, the toggle_builtin_tools tool modifies project config files, requiring careful permission assessment. Security scored 8.
Maintenance 10/20
Repository is active with 1841 stars and 25 open issues, but no clear release cadence or dependency update strategy, maintenance scored 10.
Documentation 12/20
README is detailed with installation, configuration, tool parameters, but lacks troubleshooting and cost details, documentation scored 12.
Setup experience 10/20
Provides one-line install commands and client config examples, but depends on uvx and git install, may be less friendly for Windows users, setup scored 10.

Static review · not runListed 2026-08-07

Read the FMRS scoring method →

Fit and risk

What it can accessWrites / deletes local filesUses the network

Best for

  • Claude Code users who need real-time search capabilities
  • Developers looking to reduce LLM hallucinations and get up-to-date information
  • Researchers needing web scraping and site mapping features

Not for

  • Users without a Grok API key (required)
  • Users who only need simple web scraping without configuring extra services
  • Privacy-sensitive users who do not want external APIs involved

Required permissions

  • Requires Grok API key and optionally Tavily API key
  • Requires network access to call external APIs
  • Can modify project-level .claude/settings.json file (for disabling built-in tools)

Risks and side effects

  • Depends on third-party services (Grok, Tavily), may incur API costs
  • Grok search may return inaccurate information; manual verification needed
  • Auto-disabling built-in tools may affect other functionality

Setup

Before you start

Runtime：Python 3.10+ · Prebuilt binary

GROK_API_URL required Required; Grok API endpoint in OpenAI-compatible format (supports /chat/completions and /models).
GROK_API_KEY requiredsecret Required; Grok API key from your Grok provider or mirror site.
TAVILY_API_KEY optionalsecret Optional; Tavily API key from tavily.com, used for web_fetch / web_map.
FIRECRAWL_API_KEY optionalsecret Optional; Firecrawl API key used as fallback scraping when Tavily fails.
Other optional settings (10)
GROK_MODEL optional Optional; default model, defaults to grok-4-fast, takes precedence over the config file when set.
TAVILY_API_URL optional Optional; Tavily API endpoint, defaults to https://api.tavily.com.
TAVILY_ENABLED optional Optional; whether Tavily is enabled, defaults to true.
FIRECRAWL_API_URL optional Optional; Firecrawl API endpoint, defaults to https://api.firecrawl.dev/v2.
GROK_DEBUG optional Optional; debug mode switch, defaults to false.
GROK_LOG_LEVEL optional Optional; log level, defaults to INFO.
GROK_LOG_DIR optional Optional; log directory, defaults to logs.
GROK_RETRY_MAX_ATTEMPTS optional Optional; max retry attempts, defaults to 3.
GROK_RETRY_MULTIPLIER optional Optional; retry backoff multiplier, defaults to 1.
GROK_RETRY_MAX_WAIT optional Optional; max retry wait in seconds, defaults to 10.
  1. Install Python 3.10+ and uv.
  2. Use the claude mcp add-json command to add the MCP server, configuring environment variables (GROK_API_URL and GROK_API_KEY are required, Tavily is optional).
  3. Verify installation: run claude mcp list to check connection status.
claude_desktop_config.json 
{
  "type": "stdio",
  "command": "uvx",
  "args": [
    "--from",
    "git+https://github.com/GuDaStudio/GrokSearch@grok-with-tavily",
    "grok-search"
  ],
  "env": {
    "GROK_API_URL": "https://your-api-endpoint.com/v1",
    "GROK_API_KEY": "your-grok-api-key",
    "TAVILY_API_KEY": "tvly-your-tavily-key",
    "TAVILY_API_URL": "https://api.tavily.com"
  }
}

Shown for Claude Desktop. Other clients may use a different file or key (VS Code uses "servers") — the configurator below converts it.

Generate and validate client config

Check that it works

Run claude mcp list to confirm grok-search connects, then in the conversation say "show grok-search config info" — get_config_info will test the Grok API connection and report response time and available models.

Troubleshooting

  1. 1. Ensure GROK_API_URL and GROK_API_KEY are correctly set and the API is OpenAI-compatible.
  2. 2. If encountering SSL errors, add --native-tls to uvx arguments.
  3. 3. Use the get_config_info tool to check configuration status and test connectivity.
  4. 4. Check logs (default directory: logs) for error details.

Things to try

Once connected, you can ask your AI assistant things like:

  • Search with grok-search for the latest middleware usage in the FastAPI official docs
  • Get the sources for the session_id returned by the last web_search
  • Fetch the page content of https://docs.astral.sh/uv/ and return it as Markdown
  • Call grok-search toggle_builtin_tools to disable Claude Code's built-in WebSearch and WebFetch

Tools 8

web_search read-only
Execute AI-powered web search via Grok API, with automatic time-context injection for time-sensitive queries.
get_sources read-only
Retrieve all sources for a given session_id from a previous web_search call.
web_fetch read-only
Fetch full web page content via Tavily Extract API, returning Markdown format, with automatic fallback to Firecrawl on failure.
web_map read-only
Map website structure via Tavily Map API, discovering URLs and generating a sitemap.
get_config_info read-only
Display configuration status, test Grok API connectivity, and return response time and available models.
switch_model writes
Switch the Grok model, persisting configuration to ~/.config/grok-search/config.json.
toggle_builtin_tools writes
One-click disable or enable Claude Code's built-in WebSearch and WebFetch tools.
search_planning read-only
Provide a structured search planning scaffold for generating executable search plans before complex searches.

Use cases

Get real-time web information in Claude Code, such as latest news or documentation updates.
Enable LLMs to proactively call search tools to fetch official docs, reducing hallucinations.
Fetch specific web page content and convert it to Markdown for model consumption.
Map website structure to help models understand site layout.

Supported clients

Claude Desktop
Cherry Studio

Listed from the project's documentation, not tested by this site.

Overview

Grok Search MCP is an MCP server built on FastMCP with a dual-engine architecture: Grok handles AI-driven smart search, while Tavily handles high-fidelity web scraping and site mapping. Together, they provide comprehensive real-time web access for LLM clients like Claude Code and Cherry Studio.

Similar servers

Telegram MCP Server 70 · B

Telegram MCP server powered by Telethon to let MCP clients read chats, manage groups, and send/modify messages, media, contacts, and settings.

★ 1.8k · Tools 76 Compare with this →
Compare with other MCP servers

Source revision afcdbcc68822 Data synced 2026-10-12 Read the FMRS scoring method