searxNcrawl
This MCP server gives AI agents web search plus clean Markdown/JSON page and site crawling on your own hardware.
Search the web via configured SearXNG and/or You.com providers in parallel, with merged/deduplicated results, provider attribution, and filters for language, time range, safe search, paging, engines/categories, and provider subset.
Crawl one or many URLs into Markdown or JSON, with timeouts, concurrency, exact/off deduplication, optional link removal, and Playwright
storage_statefor authenticated pages.Crawl an entire site from a seed URL using DFS within depth, page, timeout, and subdomain limits, returning Markdown or JSON.
Optionally include page content in search by crawling each result and adding its Markdown or
crawl_error.Use it via MCP tools (
search,crawl,crawl_site) or through the project's CLI/Python API.
Integrates with SearXNG to provide privacy-respecting web search capabilities, allowing users to search the web with configurable language, time range, categories, engines, and safe search.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@searxNcrawl search for latest AI trends"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
searxNcrawl is an open-source MCP server and CLI toolkit that gives AI agents web search and clean Markdown crawling, built on Crawl4AI. Search queries the configured providers, SearXNG and/or You.com, in parallel and merges the results. It runs on your own hardware, needs no API key with SearXNG and succeeds searxng-mcp.
SEARXNG_URL=http://your-searxng:8888 uvx searxncrawlQuick start
Connect your MCP client directly to searxNcrawl over STDIO or its HTTP endpoint.
You need uv and, for the search tool, at least one search provider: a SearXNG instance with JSON output enabled and/or a You.com API key (YDC_API_KEY). No SearXNG yet? Start one with Docker. Crawling works without either.
Add it to your agent
One step for every agent on your machine
uvx searxncrawl setupsetup detects Claude Code, Codex, Gemini CLI, Cursor, VS Code and Claude Desktop, asks for your SearXNG URL and, optionally, a You.com API key (hidden input, never printed), shows each config change and applies it only after you confirm. Every file it touches is backed up first, and setup --remove undoes it. More options are in MCP server.
Or per agent
The buttons and commands below assume SearXNG at http://localhost:8888; change SEARXNG_URL if yours runs elsewhere, or add YDC_API_KEY for You.com.
Claude Code
claude mcp add searxncrawl --env SEARXNG_URL=http://localhost:8888 -- uvx searxncrawlCodex
codex mcp add searxncrawl --env SEARXNG_URL=http://localhost:8888 -- uvx searxncrawlVS Code
code --add-mcp '{"name":"searxncrawl","command":"uvx","args":["searxncrawl"],"env":{"SEARXNG_URL":"http://localhost:8888"}}'Cursor (~/.cursor/mcp.json), Claude Desktop and other clients that use the mcpServers format
{
"mcpServers": {
"searxncrawl": {
"command": "uvx",
"args": ["searxncrawl"],
"env": { "SEARXNG_URL": "http://localhost:8888" }
}
}
}The first crawl downloads Playwright's Chromium once (about 550 MB). For Docker, HTTP transport and other setups see Installation and MCP server.
Claude Desktop bundle: download
searxncrawl-<version>.mcpbfrom GitHub Releases and open it in Claude Desktop.Smithery users: use the direct installation above or the native bundle; see the Smithery transition.
Ask your agent
Search for the Crawl4AI quick start, crawl the official page and summarise
how to run a first crawl.Your agent calls search, picks a result and calls crawl. It gets back clean Markdown instead of raw HTML.
Use the terminal
uvx --from searxncrawl crawl https://modelcontextprotocol.io/docs/getting-started/intro
SEARXNG_URL=http://localhost:8888 uvx --from searxncrawl search "model context protocol" --max-results 5
SEARXNG_URL=http://localhost:8888 uvx --from searxncrawl search "Python documentation" --max-results 3 --include-contentChoose how to connect
Mode | Setup | Hosting |
Native MCP over STDIO (default) |
| Runs alongside your MCP client |
Native MCP over HTTP |
| You operate the server; see HTTP configuration |
Arcade gateway (optional) | Separate Arcade adapter | Requires your own Playwright-compatible worker host and authenticated connection to Arcade |
Standard installations and releases do not require Arcade.
For shared HTTP access, use an authenticated TLS proxy. Enable Host/Origin checks with
FASTMCP_HTTP_HOST_ORIGIN_PROTECTION=truebefore startup; allowlists alone do not enable checks or authenticate callers.
Related MCP server: evo-scry
How it works
flowchart LR
agent["AI agent<br/>Claude Code · Codex<br/>Cursor · VS Code"]
server["searxNcrawl<br/>MCP server"]
searxng["SearXNG (your instance)<br/>and/or You.com API"]
browser["Crawl4AI +<br/>Playwright Chromium"]
results["Merged results<br/>attributed per provider"]
markdown["Deduplicated<br/>Markdown or JSON"]
agent -- "MCP · STDIO or HTTP" --> server
server -- "search" --> searxng --> results
server -- "crawl · crawl_site" --> browser --> markdownResults return to the agent as the tool response. Search and crawling are also available as the search and crawl CLI commands, crawling also as a Python API.
searxNcrawl exposes three MCP tools:
Tool | What it does |
| Queries the configured providers (SearXNG, You.com) in parallel and returns one merged, deduplicated list with the source of each result. Filters: language, time range, safe search, paging; categories and engines for SearXNG. With |
| Renders one or many URLs in Chromium and returns Markdown or JSON with metadata and references. |
| Follows links from a seed URL within depth, page and timeout limits, tuned for documentation sites. |
Both crawl tools drop sections that repeat within a page's Markdown (dedup_mode="exact"), can strip links for a smaller context (remove_links), and accept a Playwright storage_state for pages behind a login. Full parameters are in the MCP tools reference.
Private SearXNG instances
SEARXNG_EXTRA_HEADERSadds headers to SearXNG requests in both CLI and MCP, including access-gateway credentials and a custom User-Agent.Values such as
@~/.config/searxncrawl/cf_secretload secrets from files; absolute and~/paths are supported.Invalid header configuration disables SearXNG with a value-free warning; a configured You.com provider remains available.
Setup and JSON examples: SearXNG configuration.
Why searxNcrawl
Hosted search and crawl APIs are quick to start with, but every call goes through someone else's service and is billed per request or credit. searxNcrawl runs on your own machine or server.
searxNcrawl | ||||
Search and crawl run on your hardware | ✅ Python package or Docker; You.com queries go to You.com | ✅ Self-host stack (Docker Compose), or hosted API | ❌ Hosted API | ❌ Hosted API |
API key | Not needed with SearXNG; You.com needs one | Hosted API: needed. Self-hosted: not by default | Needed | Optional, anonymous use is rate-limited |
Cost per call | None with SearXNG; You.com per its plans | Hosted: free monthly credits, then paid credits. Self-hosted: none | Free monthly credits, then paid credits | Free monthly credits, then paid per request |
Search backend | Your SearXNG (the engines you enable) and/or You.com | Hosted: Firecrawl's search. Self-hosted: SearXNG if configured, otherwise DuckDuckGo | Tavily's search API | Exa's search index |
Licence | MIT | AGPL-3.0 | MCP server MIT, API proprietary | MCP server MIT, API proprietary |
Self-hosted Firecrawl can search through SearXNG too; it runs as a multi-service stack (API, Playwright service, Redis, RabbitMQ, PostgreSQL and more), while searxNcrawl is a single Python process with Chromium, plus your SearXNG instance if you use one.
Where the others are a better fit: if you do not want to run a search provider and a browser yourself, a hosted API is less work, and Exa's semantic search index or Firecrawl's extraction features go beyond what searxNcrawl does today. Comparison based on each project's public repository and documentation in October 2026; corrections are welcome. Firecrawl is a trademark of its owner; searxNcrawl is not affiliated with it.
FAQ
Only for the search tool, and only if you do not use You.com (YDC_API_KEY). crawl and crawl_site work without any search provider. SearXNG serves HTML only by default, so JSON output must be enabled (search.formats in settings.yml). Public instances may not enable it and often rate-limit automated requests, so run your own. Installation has a Docker one-liner.
Yes. Pages are rendered in Playwright's Chromium through Crawl4AI before they are converted to Markdown, so client-side content is captured.
Yes. Capture a session once with crawl-capture, then pass the saved storage_state file to crawl or crawl_site. See Session capture.
Start with
--transport http(port 8000 by default) or Docker Compose (port 9555).Put shared access behind an authenticated TLS proxy and connect clients to its
/mcpendpoint.Enable header protection and configure the externally visible Host and Origin allowlists.
The first crawl downloads Playwright's Chromium (about 550 MB on disk). Later runs reuse it. Set PLAYWRIGHT_AUTO_INSTALL=false where browsers are provisioned separately.
searxng-mcp is deprecated in favour of searxNcrawl. It only searched SearXNG and could pick a random public instance; searxNcrawl adds page and site crawling but needs your own SEARXNG_URL (or a You.com key), as random public-instance selection is not included.
No. Connect directly over STDIO or HTTP for the standard setup.
The optional adapter connects a separately hosted worker to Arcade's authentication and gateway.
Arcade Cloud's current runner cannot install Playwright; the worker needs a compatible host.
Documentation
Guide | Contents |
uvx, Docker Compose, pip, uv, search providers and environment variables | |
STDIO and HTTP transports, client configuration, Host and CORS guards | |
Parameters for | |
| |
| |
Markdown and JSON structure, | |
Authenticated crawling with a saved browser session | |
| |
Shared authentication and routing through Arcade; separate deployment | |
Modules and features for contributors |
Contributing
Bug reports, feature ideas and pull requests are welcome. Read the contributing guide and open an issue first for larger changes. Coding agents find project conventions in AGENTS.md. Report vulnerabilities privately as described in the security policy.
License
MIT — © 2026 DDM – Das Digitale Momentum GmbH & Co KG
Available Tools
3 toolscrawlCrawlC
Crawl one or more URLs and extract content as markdown or JSON.
Inputs: urls (required), timeout, dedup_mode, concurrency, +3 more.
crawl_siteCrawl SiteC
Crawl an entire website starting from a seed URL using DFS strategy.
Inputs: url (required), timeout, max_depth, max_pages, +5 more.
searchSearchA
Search the web via the configured search providers (SearXNG, You.com).
All configured providers are queried in parallel. Their results are merged (URL dedup, round-robin, results found by several providers first) and each result lists the providers that returned it.
Inputs: query (required), pageno, engines, language, +7 more.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 3 tools
The three tools have mostly distinct roles: search queries providers, crawl fetches specific URLs, and crawl_site traverses a whole site. The crawl vs. crawl_site boundary is clarified by the DFS/seed-URL description, though some overlap remains because crawl accepts multiple URLs.
All names are lowercase snake_case and action-oriented, which is readable and predictable. However, search and crawl are bare verbs while crawl_site uses a verb_noun compound, so the convention is not perfectly uniform.
Three tools are well scoped for a search-and-crawl server: one search entry point, one targeted URL crawler, and one whole-site crawler. Each tool earns its place without obvious redundancy.
The surface covers the core domain lifecycle: searching configured providers, extracting content from specific URLs, and crawling an entire site from a seed URL. No obvious missing operation prevents agents from accomplishing search-and-crawl workflows.
Maintenance
Related MCP Connectors
One MCP server for all web data. Scrape, search, crawl, and AI research in one install.
Web search API and MCP server for developers building AI agents.
Hosted Web Crawler MCP server for Claude, ChatGPT and Cursor, with each action switched per account.
Related MCP Servers
- AlicenseNot gradedqualityCmaintenanceMCP server for web search and content extraction using DuckDuckGo or SearXNG, with Playwright-based fetching and LLM-powered data extraction.138MIT
- AlicenseNot gradedqualityDmaintenanceMCP server for internet search via direct Google and DuckDuckGo HTML scraping with AI-powered result normalization and optional summarization, requiring no API keys for search.MIT
- AlicenseAqualityDmaintenanceA privacy-focused web search and content extraction MCP server. It integrates SearxNG with fallback to Google scraping, featuring relevance ranking, security-aware search, and rate limiting.3MIT
- AlicenseAqualityCmaintenanceMCP server for local web search via SearXNG, providing unlimited queries without API keys or cost, with automatic fallback to public instances.31MIT