grok-mcp-server
Search X (formerly Twitter) in real-time from your AI assistant using xAI's Grok API, with no X API account required.
README
grok-mcp-server
Search X (formerly Twitter) in real-time from your AI assistant — powered by xAI's Grok API.
- No X API account needed — just an xAI API key
- Quick setup —
npxwith one env var, or deploy to Vercel and connect from claude.ai - Wide compatibility — Hermes Agent, Claude Code, claude.ai, Cursor, LM Studio, and other MCP clients
MCP Client ──stdio──▶ grok-mcp-server (npx) ──API──▶ xAI Grok API
claude.ai ──OAuth 2.1──▶ grok-mcp-server (Vercel) ──API──▶ xAI Grok API
What You Can Do
- Search recent posts — Find what people are saying about any topic right now
- Summarize discussions — Get AI-curated summaries of trends, opinions, and reactions
- Extract structured data — Define a JSON Schema to pull exactly what you need (sentiment, topics, key quotes)
- Drill down with follow-ups — Chain searches to filter, compare, or dig deeper into results
- Analyze threads — Pass a post URL to get the original post and top replies ranked by engagement
Quick Start
Option A: Use via npx (Hermes Agent, Claude Code, LM Studio, Cursor, etc.)
No installation required. Add to your MCP client configuration:
{
"mcpServers": {
"grok-mcp-server": {
"command": "npx",
"args": ["-y", "grok-mcp-server@latest"],
"env": {
"XAI_API_KEY": "your-xai-api-key"
}
}
}
}
For Hermes Agent, add the server under mcp_servers in ~/.hermes/config.yaml:
mcp_servers:
grok-mcp-server:
command: "npx"
args: ["-y", "grok-mcp-server@latest"]
env:
XAI_API_KEY: "your-xai-api-key"
Restart Hermes after editing the config. The tool will be registered as mcp_grok_mcp_server_x_search.
Get your API key at console.x.ai.
Try it: "Use x_search to find recent posts about MCP servers and summarize the top opinions"
Option B: Deploy to Vercel (claude.ai web / Claude Code)
1. Deploy to Vercel
Click the button above. You'll be prompted to set environment variables — see Environment Variables for what to enter.
After deployment, open your project's root URL to verify the setup. The setup dashboard shows the status of each variable, provides a JWT secret generator, and guides you through any remaining steps.
2. Connect from claude.ai
- Log in to claude.ai
- Go to Settings > Integrations > Add More
- Enter your MCP endpoint URL:
https://your-project.vercel.app/api/mcp - Click Add — the OAuth authorization flow will start
- Enter the password you set in
AUTHORIZE_PASSWORDand click Allow - The
x_searchtool is now available in your chats!
Once registered on claude.ai, the remote MCP server is also available from Claude Code (CLI / IDE extensions) under the same account.
Try it: "Search X for reactions to the latest Grok release and return JSON with positive and negative themes"
Why grok-mcp-server?
Best for: real-time search, trend analysis, sentiment extraction, topic summarization — any read-only X research workflow.
Not for: posting, retweeting, DMs, follows, or any write operation. Use the official X MCP server (xmcp) for those.
Compared to the official X API / xmcp:
| grok-mcp-server | Official X API (xmcp) | |
|---|---|---|
| X API account | Not required | Required (pay-per-use credits) |
| Search results | AI-interpreted summaries & analysis | Raw API data |
| Structured output | Any JSON Schema | Not built-in |
| Full archive search | Available (via Grok) | Available (pay-per-use) |
| Setup | npx + 1 env var, or Vercel deploy |
Local server + X Developer app + OAuth callback setup |
| Write operations | Not supported | Supported |
For cost details, see Pricing below.
Pricing
Each x_search call incurs xAI API costs: token fees + X Search tool fee ($0.005/call).
| Model | Input | Output |
|---|---|---|
| grok-4.20-non-reasoning (default) | $1.25 / 1M tokens | $2.50 / 1M tokens |
| grok-4.3 | $1.25 / 1M tokens | $2.50 / 1M tokens |
See xAI Models and Pricing for the latest rates including cached input pricing.
Cost advantage over direct X API
The X API bills per resource ($0.005 per post returned, pay-per-usage). x_search bills per tool call — flat, regardless of how many posts Grok reads internally to answer. Concrete comparison:
| Workload | grok-mcp-server | X API direct |
|---|---|---|
| Single post + 50-reply thread | ~$0.01 (1 tool call + tokens) | ~$0.255 (1 request, 51 posts × $0.005) |
| Trend search returning 100 posts | ~$0.01 | ~$0.50 (1 request, 100 posts × $0.005) |
You also get AI-interpreted output — summaries, ranked replies, structured fields per output_schema — instead of raw JSON that you would need to summarize yourself with another LLM call.
Environment Variables
| Variable | Required | Description |
|---|---|---|
JWT_SECRET |
Yes | JWT signing key (generate via the setup dashboard or openssl rand -base64 32) |
XAI_API_KEY |
Yes | xAI API key (obtain from console.x.ai) |
AUTHORIZE_PASSWORD |
Yes | Password for the authorization screen (authorization is blocked if unset) |
BASE_URL |
No | Public URL override. Auto-detected from Vercel environment variables (VERCEL_PROJECT_PRODUCTION_URL / VERCEL_URL). Defaults to http://localhost:3000 locally. |
Local Development
git clone https://github.com/valda/grok-mcp-server.git
cd grok-mcp-server
npm install
cp .env.local.example .env.local # edit with your values
npm run dev
The dev server starts at http://localhost:3000.
CLI (stdio) development
npm run build:cli # Build dist/cli.js
npm test # Run all tests
Test the stdio server locally:
XAI_API_KEY=your-key node dist/cli.js
Architecture
- OAuth 2.1 — PKCE required, Dynamic Client Registration supported
- MCP — POST-only JSON-RPC endpoint (exposes the
x_searchtool) - Stateless — Authorization codes, access tokens, and client registrations are all signed JWTs;
Mcp-Session-Idreuses the access token (no separate session JWT)
Endpoints
| Path | Method | Description |
|---|---|---|
/ |
GET | Setup dashboard |
/.well-known/oauth-authorization-server |
GET | OAuth metadata |
/api/oauth/register |
POST | Client registration |
/api/oauth/authorize |
GET/POST | Authorization (consent screen and code issuance) |
/api/oauth/token |
POST | Token issuance (PKCE verification) |
/api/mcp |
POST | MCP JSON-RPC endpoint |
Tech Stack
- Next.js 16 (App Router)
- jose — JWT signing and verification
- @modelcontextprotocol/sdk — MCP SDK (stdio transport)
- zod — Schema validation
- tsup — CLI bundler
- Vitest — Unit and integration tests
- TypeScript 5
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。