Kindly Web Search MCP Server
A web search MCP server that gives the agent full page text at once: Stack Overflow answers, GitHub threads, arXiv papers
- 402 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The browser opens arbitrary pages from search results, and their text enters the agent context as untrusted data
- Queries go to the chosen search provider together with the API key, and with a token the server calls GitHub
shelpuk-ai-technology-consulting/kindly-web-search-mcp-server
Install
- Works in
Manual install
claude mcp add --transport stdio kindly-web-search \
-e SERPER_API_KEY="$SERPER_API_KEY" \
-e GITHUB_TOKEN="$GITHUB_TOKEN" \
-e KINDLY_BROWSER_EXECUTABLE_PATH="$KINDLY_BROWSER_EXECUTABLE_PATH" \
-- uvx --from git+https://github.com/Shelpuk-AI-Technology-Consulting/kindly-web-search-mcp-server \
kindly-web-search-mcp-server start-mcp-serverAnother provider key or SEARXNG_BASE_URL can replace SERPER_API_KEY. The server name goes right after --transport stdio.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The browser opens arbitrary pages from search results, and their text enters the agent context as untrusted data
- Queries go to the chosen search provider together with the API key, and with a token the server calls GitHub
This is third-party code. Review the repository files before installing.
What it does
Kindly Web Search searches the web and returns the content of the found pages as Markdown in the same response. The web_search tool takes a query and a result count, three by default, and returns the title, link, snippet and page_content. The get_content tool pulls the text from a given link. The agent needs no second scraper call: the question, answers and comments arrive together. The server routes each link through a chain of specialized handlers. For Stack Exchange it calls the site API, for GitHub Issues and Discussions it uses GraphQL, for Wikipedia it queries MediaWiki, and arXiv papers come through the Atom API with the PDF converted to Markdown. Every other page opens in a headless Chromium-based browser through nodriver. Documentation sites that serve Markdown have a fast path: the server first tries to download ready Markdown with a plain request and starts the browser only on a miss. There are many search providers: Serper, SerpBase, Tavily, SearXNG, Sofya, You.com, Serply and apifare. The server uses the first one whose key is set, in that order.
Who it is for. For developers who debug errors in Claude Code, Codex or Cursor and want the agent to receive whole discussions together with search results.
Good fit when
- An agent is debugging a rare error and needs the Stack Overflow question with its accepted answers and comments
- You need a fix from the Issues or Discussions of a GitHub repository, with reactions and thread replies
- You need fresh documentation pages of a recently released library: the server opens any page in a browser
- A conclusion needs the text of a paper: an arXiv link becomes Markdown including the PDF content
- Search must go through your own SearXNG instance on your server
Not a fit when
- You have no search provider key and no SearXNG of your own: search will not work without them
- The machine has no Chromium-based browser: text extraction may fail for arbitrary sites, and Stack Exchange, GitHub, Wikipedia and arXiv keep working
- You need search over closed sources and corporate wikis: the server works with open pages of the web
- You only need a list of links and snippets: full page text then just spends the agent's context
Example request
Find why a Google Cloud Batch job with a GPU fails after half an hour, and read the discussions you find in fullMCP
- Transport
- stdio
- Authentication
- API key
Environment variables
SERPER_API_KEYrequired, secret- Serper key, the primary search provider; another provider key or SEARXNG_BASE_URL can replace it
GITHUB_TOKENsecret- Read-only GitHub token for public repositories, improves Issues output
KINDLY_BROWSER_EXECUTABLE_PATH- Path to a Chromium-based browser when auto-detection misses it
SEARXNG_BASE_URL- Address of your own SearXNG instance instead of a commercial provider
Limitations
You need a key for one of the search providers or a SearXNG address, and for arbitrary sites also a Chromium-based browser on the same machine as the MCP client: Chrome, Chromium, Edge or Brave. The browser path can be set with the KINDLY_BROWSER_EXECUTABLE_PATH variable.
GITHUB_TOKEN is optional and makes Issues render in a model-friendly form and hit rate limits less often. A token with read-only access to public repositories is enough.
Python 3.13 and 3.14 are supported. On 3.14 the advanced PDF layout extras are disabled because onnxruntime wheels may be unavailable. Whether the search providers are reachable from Russia was not checked for this card.
How to disable. Run `claude mcp remove kindly-web-search` in Claude Code or `codex mcp remove kindly-web-search` in Codex. If the server is set in a file, delete the kindly-web-search entry from .mcp.json, ~/.codex/config.toml or ~/.gemini/settings.json. Revoke the provider keys in their dashboards.
FAQ
How do I install Kindly Web Search in Claude Code?
Install uv, set the key of one search provider such as `SERPER_API_KEY`, and run `claude mcp add --transport stdio kindly-web-search` with `-e` flags for the variables. After the `--` separator comes the launch through `uvx --from git+https://github.com/Shelpuk-AI-Technology-Consulting/kindly-web-search-mcp-server kindly-web-search-mcp-server start-mcp-server`. The server name must sit right after `--transport stdio` and before the `-e` flags.
How does Kindly differ from an ordinary search MCP?
Ordinary servers return a title, link and snippet, and the agent fetches the page text in a second call. Kindly puts the Markdown of the found page into page_content right away. For Stack Overflow, GitHub Issues and Discussions, Wikipedia and arXiv it uses their APIs, so the response includes answers, comments and reactions.
Is a browser required, and can I go without one?
Without a Chromium browser the specialized sources still work: Stack Exchange, GitHub, Wikipedia and arXiv. Other sites need Chrome, Chromium, Edge or Brave on the same machine as the client. If auto-detection misses the browser, set the path in KINDLY_BROWSER_EXECUTABLE_PATH.
How do I use my own SearXNG instead of a commercial search?
Set SEARXNG_BASE_URL to the address of your instance. If it needs authentication or blocks bots, add SEARXNG_HEADERS_JSON and SEARXNG_USER_AGENT. The selection order is SERPER_API_KEY, SERPBASE_API_KEY, TAVILY_API_KEY, then SearXNG, then the rest.
What if Claude Code times out while starting the server?
The first uvx call builds the environment for 30–60 seconds. Run the launch command once in a terminal to warm it up, then reconnect. For Claude Code the project suggests raising the startup timeout to 120 seconds.
Discussion
No questions or reviews yet.
Related
MCP serversAn MCP server for private web search through your SearXNG instance: search with pagination and filters, page and PDF reading
MCP server for real-time web search: general and vertical search, batch queries and page content extraction
Web search via DuckDuckGo and page reading for the agent with no API keys, with built-in rate limiting
A unified search MCP server: one interface to Tavily, Brave, Kagi, Exa, Linkup, Firecrawl and GitHub search
Official SerpApi MCP server: the agent gets results from Google, Bing, YouTube, Amazon and other engines as JSON or Markdown
Yandex and Google search MCP server
web-compass
An MCP web-search and content-extraction server via Yandex and Google on top of SerpAPI, geared toward RU/CIS results
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Kindly Web Search MCP Server (this entry) | no | API key | 402 | |
| open-webSearch | no | not required | 1,880 | key-free search by parsing search engine pages, without full-text extraction |
| Brave Search MCP Server | yes | API key | 1,544 | Brave's official server for web, news, image and place search |
| AnySearch MCP Server | no | key is optional | 2,052 | a real-time search server with batch queries and vertical sources |
README in short
The project is part of the Shelpuk AI Technology Consulting suite, as one of three MCP servers (Serena, Kindly Web Search, Lad MCP Server) alongside the `tdd` skill. Separate connection recipes cover Codex, Claude Code, Gemini CLI, OpenClaw and Antigravity, including PowerShell variants. Extra variables tune the Markdown fast path and GitHub Issues output.