使用说明
Grok API MCP公司
两个独立的工具用于Grok与web和X(Twitter)搜索基础的有状态交互:
- CLI客户端(Rust) -独立
grok-ask直接调用Grok API的二进制文件 - MCP服务器(Python) -FastMCP后端用于基于工具的集成(Claude Desktop等)
特性
- 网络搜索限制:自动网络搜索当前/事实信息
- X(推特)搜索暂停:使用句柄、日期和媒体过滤器搜索X个帖子
- 有状态的对话:通过以下方式维护多个查询的上下文
response_id - 多种工具:搜索、询问、思考、聊天、x-Search和x-ask模式
建筑
更喜欢直接使用Rust CLI。当您特别需要在像Claude Desktop这样的主机中集成MCP时,请联系Python MCP服务器。
CLI和MCP服务器是相同Grok Responses API的独立客户端:
┌─────────────────────────────────────────────────────────┐
│ CLI Client (Rust) - grok-ask │
│ - Standalone binary, no dependencies │
│ - Direct Grok API calls via HTTP │
│ - Use as Claude Code skill or command line │
└─────────────────────┬───────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────┐
│ Grok API │
│ - Models: grok-4-1-fast-non-reasoning / grok-4-1-fast │
│ - Automatic grounding (web_search, x_search) │
│ - Server-side conversation state │
└─────────────────────┬───────────────────────────────────┘
▲
│
┌─────────────────────┴───────────────────────────────────┐
│ MCP Server (Python/FastMCP) - optional │
│ - For Claude Desktop, Cursor, etc. │
│ - Same API features via MCP tools │
└─────────────────────────────────────────────────────────┘
快速开始
命令行用法
# Install the binary first
cd cli
cargo install --path .
# Make sure Cargo binaries are on your PATH
export PATH="$HOME/.cargo/bin:$PATH"
# Set API key
export XAI_API_KEY=your_key_here
# Quick web search
grok-ask --search "latest AI news"
# Get grounded answer
grok-ask --ask "What is xAI?"
# Deep reasoning
grok-ask --think "Compare Grok to GPT-4"
# Chat without web search
grok-ask --chat "Tell me a joke"
# Search X (Twitter) posts
grok-ask --x-search "AI news"
# Get answers grounded in X posts
grok-ask --x-ask "What are people saying about Grok?"
# X search with filters
grok-ask --x-search "AI safety" --allowed-handles elonmusk,sama
grok-ask --x-search "SpaceX" --from-date 2025-01-01 --to-date 2025-01-15
grok-ask --x-ask "Latest xAI updates" --enable-images --enable-video
# Follow-up conversation
grok-ask --ask "What products does xAI offer?" -r <response_id>
# JSON output
grok-ask --ask "Query" -o json
MCP服务器使用情况
cd server
uv sync
cp .env.example .env
# Edit .env with your XAI_API_KEY
uv run python server.py
然后使用工具通过MCP客户端连接: search, ask, think, chat, x_search, x_ask
安装
CLI(Rust)
# Install globally
cd cli
cargo install --path .
# If grok-ask is not found afterwards, add Cargo's bin directory to your shell PATH
echo 'export PATH="$HOME/.cargo/bin:$PATH"' >> ~/.zshrc
# Reload your shell if you want the new PATH to apply immediately in the current terminal.
# This is only needed after editing ~/.zshrc; opening a new terminal works too.
source ~/.zshrc
# Or build locally
cargo build --release
# Binary at: ./target/release/grok-ask
MCP服务器(Python)
cd server
uv sync
cp .env.example .env
# Edit .env with your XAI_API_KEY
API密钥
从以下位置获取Grok API密钥:https://console.x.ai/
工具
| 工具 | 模型 | 搜索 | 最大令牌 |
|---|---|---|---|
search |
grok-4-1快速推理 | 网络 | 4096 |
ask |
grok-4-1快速推理 | 网络 | 8192 |
think |
grok-4-1-fast | 网络 | 16384 |
chat |
grok-4-1不讲道理 | 无 | 8192 |
x_search |
grok-4-1快速推理 | X(推特) | 4096 |
x_ask |
grok-4-1快速推理 | X(推特) | 8192 |
X搜索筛选器选项
| 选项 | 描述 |
|---|---|
allowed_handles |
仅包括来自这些X手柄的帖子(最多10个) |
excluded_handles |
从这些X手柄中排除帖子(最多10个) |
from_date |
YYYY-MM-DD格式的开始日期 |
to_date |
YYYY-MM-DD格式的结束日期 |
enable_images |
允许模型分析帖子中的图像 |
enable_video |
允许模型分析帖子中的视频 |
用法示例
搜索当前信息
search("latest AI news 2025")
问一个事实问题
ask("What is xAI and what products do they offer?")
深度分析
think("Compare Grok's capabilities to other AI models")
普通聊天(无网络搜索)
chat("Tell me a joke about programming")
搜索X个帖子
x_search("AI safety discussions")
使用筛选器搜索X
# Only from specific handles
x_search("AI news", allowed_handles=["elonmusk", "sama"])
# Date range
x_search("SpaceX launch", from_date="2025-01-01", to_date="2025-01-15")
# With media understanding
x_search("xAI demos", enable_images=True, enable_video=True)
从X个帖子中获取答案
x_ask("What are people saying about Grok?")
# From specific accounts
x_ask("Latest xAI announcements", allowed_handles=["xai"])
多回合对话
# First question
response = ask("What is quantum computing?")
# Returns: ... response_id: resp_abc123
# Follow-up (Grok remembers context)
response = ask("How is it different from classical computing?", response_id="resp_abc123")
有状态的对话
API通过 response_id:
- 无需重新发送对话历史记录
- 隐式缓存(更快、更便宜)
- 通过
response_id从之前的回复继续对话
Claude桌面集成
添加到您的Claude Desktop配置(~/Library/Application Support/Claude/claude_desktop_config.json):
{
"mcpServers": {
"grok": {
"command": "uv",
"args": [
"run",
"--directory",
"/path/to/grok-api-mcp/server",
"python",
"server.py"
],
"env": {
"XAI_API_KEY": "your_api_key_here"
}
}
}
}
项目结构
grok-api-mcp/
├── cli/
│ ├── src/
│ │ └── main.rs # Rust CLI
│ └── Cargo.toml
├── server/
│ ├── server.py # MCP server (Python/FastMCP)
│ ├── pyproject.toml
│ └── .env.example
├── skill.md # Claude Code skill
└── README.md
API 参考
CLI和MCP服务器都使用Grok API端点: https://api.x.ai/v1/responses
请求格式(Web搜索)
{
"model": "grok-4-1-fast-non-reasoning",
"input": [
{"role": "user", "content": "Your question"}
],
"tools": [
{"type": "web_search"}
]
}
请求格式(X搜索)
{
"model": "grok-4-1-fast-non-reasoning",
"input": [
{"role": "user", "content": "Your question"}
],
"tools": [
{
"type": "x_search",
"allowed_x_handles": ["elonmusk"],
"from_date": "2025-01-01",
"to_date": "2025-01-15",
"enable_image_understanding": true,
"enable_video_understanding": true
}
]
}
对于 think 命令,使用 grok-4-1-fast (推理能力模型)。
认证
通过以下方式进行承载令牌身份验证 XAI_API_KEY 环境变量。
学分
基于 双子相互作用mcp 通过 @DigiBugCat.
许可证
麻省理工学院