Works with the AI Agents you already use
Key-Free Multi-Engine Search MCP
Give your agent free web search and article reading via one self-hosted MCP server — multiple engines, no API keys, stdio, HTTP or Docker.
Free
Key-Free Multi-Engine Search MCP
Example session with this skill installed
I need my coding agent to search the web without paying for a search API. Set up a key-free aggregated search MCP server: default engine DuckDuckGo with Bing as fallback, exposing web_search plus a generic page reader, running in HTTP mode on port 3000 through Docker so both Claude Desktop on macOS and my dev box can connect. Wire up the client config JSON, set sane result caps and timeouts so we don't get rate-limited, and walk me through a smoke test — search 'vector database benchmarks', read the top result, and summarize it.
- Read your context and instructions
- Compiled the key-free multi-engine search
Summary
The key-free-search MCP server is configured for multi-engine web discovery plus clean article reading, deployed in HTTP mode on port 3000. Claude Desktop and your dev box can now search DuckDuckGo and Bing with zero API keys.
| Item | Value |
|---|---|
| Transport | streamable HTTP at http://localhost:3000/mcp |
| Engines | duckduckgo (default), bing (fallback) |
| Tools exposed | web_search, read_page |
| Result cap / cache | 20 per query / 60s TTL |
| Timeout / proxy | 15s / off |
Client config written to mcp.json:
{
"mcpServers": {
"web-search": {
"transport": { "type": "streamableHttp", "url": "http://localhost:3000/mcp" }
}
}
}
Launch command
docker run -d --name key-free-search -p 3000:3000 \
-e MODE=http -e DEFAULT_SEARCH_ENGINE=duckduckgo \
-e ALLOWED_SEARCH_ENGINES=duckduckgo,bing -e MAX_RESULTS=20 \
your-image:latest
Next steps
- Smoke test: call
web_searchwith{"query":"vector database benchmarks","limit":5}and verify every row hastitle,url,snippet,engine. - Feed one hit to
read_page(maxChars 12000) and confirm boilerplate is stripped. - Check hourly request counts in the logs; enable
USE_PROXYif an engine starts timing out. - Add
read_github_readmeorread_csdn_articlelater if code-doc or tutorial workflows appear.
Connects securely to your tools. The creator never sees your data.
About this skill
Stop paying for expensive search APIs just to give your agent internet access. Most search-enabled agents fail because they rely on single-source snippets that lack the depth for complex reasoning. This skill provides a self-hosted Model Context Protocol (MCP) server that aggregates multiple public search engines without requiring a single API key. It solves the "stale knowledge" problem by allowing agents to discover fresh links and then deeply read page content to extract specific facts.
What it does
- Multi-engine aggregation queries DuckDuckGo, Bing, Brave, and others simultaneously for broader coverage.
- Content extraction fetches clean article text while stripping navigation, footers, and cookie banners.
- Platform-specific reading includes optimized logic for GitHub READMEs and technical community posts.
- Transport flexibility supports both stdio for local CLI agents and HTTP/SSE for networked deployments.
- Result normalization converts varied engine outputs into a stable, structured schema for consistent agent parsing.
How it works
- Deployment involves running the server via Docker or
npxwith your preferred environment variables. - Discovery occurs when the agent calls
web_searchto get a list of normalized links and snippets. - Selection happens as the agent identifies the most relevant URLs based on the search metadata.
- Extraction uses the
read_pagetool to pull the full text of chosen articles into the context window.
Frameworks & tools
Built for the Model Context Protocol (MCP). Works with Node.js, Docker, and any MCP-compliant client like Claude Desktop or Cursor. Uses standard libraries for HTML parsing and HTTP networking.
Why this beats prompting it yourself
Prompting an agent to "search the web" only works if it has a pre-configured tool. This skill provides the actual infrastructure, handling the messy realities of DOM parsing, engine-specific rate limits, and result deduplication that raw prompts cannot manage.
Use cases
- Technical research finding and reading documentation across GitHub and forums.
- Market monitoring tracking recent news without hitting API credit limits.
- Competitor analysis aggregating snippets from multiple search engines for comparison.
Known limitations
Heavy reliance on public HTML makes it susceptible to breakage if search engines change their site structure. It does not support JavaScript-heavy pages requiring a headless browser.
Depends on public search HTML, so engines can break or rate-limit without notice — it is not meant for bulk harvesting. JavaScript-rendered pages may extract poorly since the fetcher sees only server-rendered HTML.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
1 use
Times used to date
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 1 month ago
- Free with an Agensi account
Requires
Permissions
Allowed Hosts
File Scopes
Needs network to query public search engines and fetch pages, plus file writes for client config and shell for npx/Docker launches.