菜单
Web search MCP server with intelligent content extraction for AI coding tools, supporting Serper, Tavily, and SearXNG with specialized parsers for StackOverflow, GitHub, Wikipedia, and arXiv.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/Aradotso/mcp-skills --skill kindly-web-search-mcp-server命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
基于 SOC 职业分类
正在显示 SKILL.md
| name | kindly-web-search-mcp-server |
| description | Web search MCP server with intelligent content extraction for AI coding tools, supporting Serper, Tavily, and SearXNG with specialized parsers for StackOverflow, GitHub, Wikipedia, and arXiv. |
| triggers | ["search the web for","find documentation about","look up recent discussions on","get content from this URL","search StackOverflow for","find GitHub issues about","retrieve article content from","search for API documentation"] |
Skill by ara.so — MCP Skills collection.
Kindly Web Search is an MCP server that provides intelligent web search and content extraction for AI coding assistants. Unlike basic search wrappers, it returns full structured content in a single call—including StackOverflow answers, GitHub issue comments, Wikipedia articles, and arXiv papers—eliminating the need for separate scraping tools.
Key differentiator: When searching for solutions, Kindly fetches the entire conversation (questions + answers + comments) in one request, not just a title and URL snippet.
Eliminates need for:
SERPER_API_KEY (recommended)TAVILY_API_KEYSEARXNG_BASE_URL (self-hosted SearXNG)GITHUB_TOKEN (read-only, public repos only)# macOS/Linux
curl -LsSf https://astral.sh/uv/install.sh | sh
# Windows (PowerShell)
irm https://astral.sh/uv/install.ps1 | iex
# macOS
brew install --cask chromium
# Ubuntu/Debian
sudo apt-get update && sudo apt-get install -y chromium
# Windows: Install Chrome or Edge from official sources
# Choose ONE search provider
export SERPER_API_KEY="your-key-here"
# OR
export TAVILY_API_KEY="your-key-here"
# OR
export SEARXNG_BASE_URL="https://searx.example.org"
# Optional: Better GitHub extraction
export GITHUB_TOKEN="your-github-token"
# Optional: Custom browser path (auto-detected by default)
export KINDLY_BROWSER_EXECUTABLE_PATH="/path/to/chrome"
Codex:
# Using Serper
codex mcp add kindly-web-search \
--env SERPER_API_KEY="$SERPER_API_KEY" \
--env GITHUB_TOKEN="$GITHUB_TOKEN" \
-- uvx --from git+https://github.com/Shelpuk-AI-Technology-Consulting/kindly-web-search-mcp-server \
kindly-web-search-mcp-server start-mcp-server
# Using Tavily
codex mcp add kindly-web-search \
--env TAVILY_API_KEY="$TAVILY_API_KEY" \
--env GITHUB_TOKEN="$GITHUB_TOKEN" \
-- uvx --from git+https://github.com/Shelpuk-AI-Technology-Consulting/kindly-web-search-mcp-server \
kindly-web-search-mcp-server start-mcp-server
Claude Desktop / Other MCP Clients:
Add to your MCP settings JSON (typically ~/Library/Application Support/Claude/claude_desktop_config.json on macOS):
{
"mcpServers": {
"kindly-web-search": {
"command": "uvx",
"args": [
"--from",
"git+https://github.com/Shelpuk-AI-Technology-Consulting/kindly-web-search-mcp-server",
"kindly-web-search-mcp-server",
"start-mcp-server"
],
"env": {
"SERPER_API_KEY": "your-key-here",
"GITHUB_TOKEN": "your-github-token"
}
}
}
}
Windows (PowerShell):
codex mcp add kindly-web-search `
--env SERPER_API_KEY="$env:SERPER_API_KEY" `
--env GITHUB_TOKEN="$env:GITHUB_TOKEN" `
-- uvx --from git+https://github.com/Shelpuk-AI-Technology-Consulting/kindly-web-search-mcp-server `
kindly-web-search-mcp-server start-mcp-server
web_searchSearch the web and return results with full page content.
Parameters:
query (string, required): Search querynum_results (integer, optional): Number of results (default: 3, max: 10)Returns:
title: Page titlelink: URLsnippet: Brief excerptpage_content: Full page content in Markdown (extracted intelligently based on source)Example usage in prompt:
Search for "FastAPI async database connection pooling best practices"
What happens behind the scenes:
# The MCP server executes:
results = web_search(
query="FastAPI async database connection pooling best practices",
num_results=5
)
# Returns structured results with full content:
# - StackOverflow threads include all answers and comments
# - GitHub issues include full conversation threads
# - Documentation pages return clean Markdown
# - arXiv papers return full text content
get_contentExtract content from a specific URL.
Parameters:
url (string, required): URL to extract content fromReturns:
page_content: Full page content in MarkdownExample usage in prompt:
Get the full content from https://github.com/tiangolo/fastapi/issues/12345
What happens behind the scenes:
# The MCP server executes:
content = get_content(url="https://github.com/tiangolo/fastapi/issues/12345")
# Returns full GitHub issue with:
# - Issue description
# - All comments
# - Reactions and metadata
# - Code snippets in proper format
Kindly automatically detects and optimally extracts content from:
Returns complete Q&A threads with:
Returns full conversation threads:
Requires GITHUB_TOKEN for best results (avoids rate limits, returns richer structure).
Returns clean article content:
Returns full paper content:
Uses headless browser (nodriver) to:
Kindly checks providers in this order:
SERPER_API_KEY set) — recommended, most reliableTAVILY_API_KEY set)SEARXNG_BASE_URL set)# Search providers (set ONE)
SERPER_API_KEY= # Google Serper API key
TAVILY_API_KEY= # Tavily API key
SEARXNG_BASE_URL= # Self-hosted SearXNG instance URL
# Content extraction
GITHUB_TOKEN= # GitHub personal access token (read-only)
KINDLY_BROWSER_EXECUTABLE_PATH= # Custom browser path (auto-detected)
# SearXNG advanced config
SEARXNG_HEADERS_JSON= # JSON string with custom headers
SEARXNG_USER_AGENT= # Custom user agent for SearXNG
Kindly auto-detects browsers in this order:
KINDLY_BROWSER_EXECUTABLE_PATH (if set)Windows manual path (if auto-detection fails):
$env:KINDLY_BROWSER_EXECUTABLE_PATH="C:\Program Files\Google\Chrome\Application\chrome.exe"
Prompt:
I'm getting "RuntimeError: CUDA out of memory" in PyTorch. Search for solutions and workarounds.
Result: Kindly searches StackOverflow, GitHub issues, and documentation, returning complete threads with:
Prompt:
Search for the latest FastAPI dependency injection documentation and examples.
Result: Returns official docs with:
Prompt:
Get the full content of the attention mechanism paper from arXiv.
Result: Returns complete paper text with:
Prompt:
Find GitHub issues about memory leaks in Langchain streaming responses.
Result: Returns relevant issues with:
Prompt:
Search for solutions to "GCP Cloud Batch fails with GPU instance template"
What you get:
# StackOverflow Result 1
## Question (Score: 15)
I am trying to run a GCP Cloud Batch job with K80 GPU...
[full question with code]
## Answer (Accepted, Score: 23)
The issue is related to GPU quota limits...
[complete solution with configuration examples]
## Comments
- User123: "This also works for T4 GPUs"
- OP: "Confirmed, this fixed it. Also needed to..."
# GitHub Issue Result 2
## googleapis/python-batch #145 - GPU instance template failures
[Full issue thread with 12 comments and resolution]
Prompt:
Search for best practices on using SQLAlchemy async sessions with FastAPI
What you get: Multiple results including:
Problem: None of the search provider env vars are set.
Solution:
export SERPER_API_KEY="your-key"
# OR
export TAVILY_API_KEY="your-key"
# OR
export SEARXNG_BASE_URL="https://searx.example.org"
Problem: Chromium-based browser not installed or not detected.
Solution:
# Install Chrome/Chromium
brew install --cask chromium # macOS
sudo apt-get install chromium # Linux
# Or set custom path
export KINDLY_BROWSER_EXECUTABLE_PATH="/usr/bin/chromium"
Problem: First uvx invocation takes 30-60 seconds to build environment.
Solution: Run once in terminal to prewarm:
uvx --from git+https://github.com/Shelpuk-AI-Technology-Consulting/kindly-web-search-mcp-server \
kindly-web-search-mcp-server start-mcp-server
Then restart your MCP client.
Problem: GitHub API returns 403 or rate limit errors.
Solution: Set GITHUB_TOKEN:
# Create token at https://github.com/settings/tokens
# Needs: public_repo (classic) or repo read access (fine-grained)
export GITHUB_TOKEN="ghp_..."
Problem: Some pages return minimal content.
Possible causes:
Workarounds:
When working with AI coding agents (Claude Code, Codex, Cursor):
Start broad: Let the agent search first
Search for FastAPI WebSocket authentication patterns
Then drill down: Use get_content for specific URLs
Get the content from [URL the agent found]
Combine with code generation: The agent now has full context
Based on the search results, implement WebSocket auth for our API
Kindly works best when combined with:
Together these improve AI-generated code quality by 15-20%.
If using self-hosted SearXNG:
export SEARXNG_BASE_URL="https://searx.example.org"
# If instance requires authentication
export SEARXNG_HEADERS_JSON='{"Authorization":"Bearer your-token"}'
# If instance blocks default user agents
export SEARXNG_USER_AGENT="Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36"
Issues and PRs welcome. See repository for contribution guidelines.