Server data from the Official MCP Registry
11-engine MCP search server for AI agents. Free, zero API keys needed. Waterfall, verify, extract.
11-engine MCP search server for AI agents. Free, zero API keys needed. Waterfall, verify, extract.
Agent Search MCP is a well-architected search aggregator with reasonable security controls and sensible permission scope for its purpose. The codebase demonstrates good engineering practices including input validation, error handling, and infrastructure components (rate limiting, health tracking, caching). However, there are moderate-severity concerns around environment variable handling in the paid API integrations, lack of explicit SSRF validation in the codebase provided, and limited documentation of credential security. The server's permissions (network_http, env_vars) are appropriate for a Developer Tools category search service. Supply chain analysis found 2 known vulnerabilities in dependencies (1 critical, 0 high severity). Package verification found 1 issue.
4 files analyzed Β· 10 issues found
Security scores are indicators to help you make informed decisions, not guarantees. Always review permissions before connecting any MCP server.
This plugin requests these system permissions. Most are normal for its category.
Set these up before or after installing:
Environment variable: BRAVE_API_KEY
Environment variable: TAVILY_API_KEY
Environment variable: EXA_API_KEY
Environment variable: MODE
Environment variable: PORT
Add this to your MCP configuration file:
{
"mcpServers": {
"io-github-lennney-agent-search-mcp": {
"env": {
"MODE": "your-mode-here",
"PORT": "your-port-here",
"EXA_API_KEY": "your-exa-api-key-here",
"BRAVE_API_KEY": "your-brave-api-key-here",
"TAVILY_API_KEY": "your-tavily-api-key-here"
},
"args": [
"-y",
"agent-search-mcp"
],
"command": "npx"
}
}
}From the project's GitHub README.
π 11 search engines, 8 free, one MCP server. Zero API keys. Chinese search. Multi-source verification. Streamable HTTP.
npm installis enough.
Works with Claude Code, Claude Desktop, Cursor, Windsurf, VS Code, Codex, Hermes, and any MCP-compatible client.
β Star on GitHub β it helps others discover the project!
English Β· δΈζ Β· Tools Β· CLI Β· Engines Β· Config
# npx (no install)
npx agent-search-mcp
# Global install
npm install -g agent-search-mcp
{
"mcpServers": {
"agent-search": {
"command": "npx",
"args": ["agent-search-mcp"]
}
}
}
{
"mcpServers": {
"agent-search": {
"command": "npx",
"args": ["agent-search-mcp"]
}
}
}
{
"mcpServers": {
"agent-search": {
"command": "npx",
"args": ["agent-search-mcp"]
}
}
}
{
"mcpServers": {
"agent-search": {
"command": "npx",
"args": ["agent-search-mcp"]
}
}
}
{
"mcpServers": {
"agent-search": {
"command": "npx",
"args": ["agent-search-mcp"]
}
}
}
mcp_servers:
agent-search:
command: npx
args: ["agent-search-mcp"]
{
"mcpServers": {
"agent-search": {
"command": "npx",
"args": ["agent-search-mcp"]
}
}
}
AI agents need to search the web. But existing options have problems:
| Solution | Price | The Catch |
|---|---|---|
| Tavily | $0.01/search | Monthly cost: $20-50+ |
| Exa | $50/mo | Powerful semantic search, but expensive |
| Brave Search | $3/1K after 2K free | Free quota runs out fast |
| DDG MCP | Free | Single source, no verification, results vary |
| Serper | $0.30/1K | Google SERP, no content extraction |
Agent Search MCP is different:
| Capability | Why It Matters |
|---|---|
| 8 free engines, zero API keys | DuckDuckGo, Sogou, Bing, Baidu, Wikipedia, Startpage, Yandex, Mojeek β all work out of the box |
| Waterfall progressive search | Runs engines in confidence-gated phases. Stops early when results are sufficient β saves 50-75% calls |
| Multi-source verification | Cross-checks results across engines. Each result scored 1-3 based on how many sources agree |
| Chinese search | Sogou + Baidu for native Chinese web search. Not a translation layer |
| Content enrichment | Auto-extracts full page content for low-confidence results via Jina Reader |
| MCP 2025 compliant | Streamable HTTP transport, readOnlyHint/idempotentHint annotations, explicit capabilities |
| Token optimized | Smart truncation + dedup saves ~40-50% tokens vs raw search |
| Self-hostable | No third-party data sharing. Run on your own VPS |
| Security built-in | SSRF protection, prompt injection detection, URL validation |
| Engine | Free | Strengths |
|---|---|---|
| DuckDuckGo | β | Privacy-focused, English web. Python ddgs preferred, cheerio HTML fallback |
| Sogou | β | Chinese web search, WeChat content |
| Bing | β | Multilingual, strong English results |
| Baidu | β | Chinese web search, Baidu Baike |
| Wikipedia | β | Clean JSON API, structured knowledge |
| Startpage | β | Google results via privacy proxy |
| Yandex | β | Russian web search |
| Mojeek | β | Independent crawler, privacy-focused |
| Brave Search | β | High-quality web results, 2K free/month |
| Tavily | β | Agent-optimized search, 1K free/month |
| Exa | β | Neural semantic search, 1K free/month |
| Tool | Description | Best For |
|---|---|---|
free_search | Multi-engine search with auto-fallback | Quick fact-finding, general queries |
free_search_advanced | Filtered search with waterfall, domain filtering, enrichment | High-confidence results, date ranges, domain filtering |
free_search_news | News search across DDG News + Bing News | Recent news, current events |
search_with_synthesis | Deep search with prompt_hint for LLM synthesis | Complex queries needing multi-source verification |
free_extract | Extract full page content as Markdown | Reading a specific page from search results |
fetch_github_readme | Fetch README from a GitHub repo | Quick project documentation |
fetch_csdn_article | Fetch content from CSDN blog | Chinese developer articles |
fetch_juejin_article | Fetch content from Juejin | Chinese developer articles |
All tools are read-only and idempotent with MCP 2025 annotations.
Waterfall Search (3 phases, confidence-gated):
Stops as soon as results are sufficient. Saves 50-75% engine calls.
Confidence Scoring (1-3):
Structured Errors: Engine failures return typed errors (timeout, rate_limited, permission_denied, etc.) with actionable suggestions β agents can self-recover.
| Variable | Default | Description |
|---|---|---|
BRAVE_API_KEY | β | Brave Search API key (2K free/month) |
TAVILY_API_KEY | β | Tavily API key (1K free/month) |
EXA_API_KEY | β | Exa API key (1K free/month) |
LOG_LEVEL | info | Log level: info, debug |
MODE | stdio | Transport: stdio, http, both |
PORT | 3000 | HTTP server port (MODE=http/both) |
Zero config works β no API keys needed for the 8 free engines.
Control which tools your agent can see:
# Only search tools
ENABLED_TOOLS=free_search,free_search_advanced,free_search_news
# Disable specific tools
DISABLED_TOOLS=free_extract,fetch_github_readme
# Enable only one fetch tool
ENABLED_TOOLS=fetch_csdn_article
| Variable | Description |
|---|---|
ENABLED_TOOLS | Comma-separated list of tools to enable. If set, only these are registered |
DISABLED_TOOLS | Comma-separated list of tools to disable. Takes priority over ENABLED_TOOLS |
# Only use Chinese engines
ALLOWED_ENGINES=sogou,baidu
# Exclude specific engines
DENIED_ENGINES=yandex,mojeek
agent-search-mcp also works as a standalone CLI tool (fasm).
# Search
fasm search "TypeScript MCP server"
fasm search "query" --count 5 --engines bing,baidu --json
# Extract
fasm extract "https://example.com"
fasm extract "https://example.com" --json
# HTTP server
fasm serve --port 8080
Agent (Claude Code, Cursor, etc.)
β MCP Protocol (stdio / Streamable HTTP)
MCP Server
βββ Tools Layer
β βββ free_search / free_search_advanced / free_search_news
β βββ search_with_synthesis
β βββ free_extract
β βββ fetch_github_readme / fetch_csdn_article / fetch_juejin_article
βββ Aggregation Layer
β βββ Waterfall Search (3-phase confidence-gated)
β βββ Cross-Engine Scoring (frequency + domain authority)
β βββ Dedup (URL + title)
β βββ Content Enrichment (Jina Reader)
β βββ Query Expansion (rule-based, 4 strategies)
βββ Engine Layer (11 engines)
β βββ Free: DDG, Sogou, Bing, Baidu, Wikipedia, Startpage, Yandex, Mojeek
β βββ Paid: Brave, Tavily, Exa
βββ Infrastructure
βββ Health Tracker (per-engine circuit breaking)
βββ Rate Limiter (adaptive concurrency)
βββ Cache (LRU, 60s TTL, 1000 entries)
βββ SSRF Protection (URL validation)
βββ Security (injection detection, boundary markers)
git clone https://github.com/lennney/agent-search-mcp.git
cd agent-search-mcp
npm install
npm run build
npm test # 448 tests, 40 files
npm run dev # stdio mode
npm run dev:http # HTTP mode (port 3000)
| Metric | Value |
|---|---|
| Tests | 448 passing, 40 files |
| Engines | 11 (8 free, 3 paid) |
| MCP Tools | 8 |
| Dependencies | 5 production |
| Node.js | >= 18 |
π‘οΈ mcp-slim-guard β Add security + compression to your MCP stack
npm install -g mcp-slim-guard
mcp-slim-guard init
mcp-slim-guard start
Pair agent-search-mcp with mcp-slim-guard to get:
| Feature | Benefit |
|---|---|
| Schema compression | Reclaim ~83% of context window β 1,736 β 300 tokens |
| Tool allow/deny | Glob-based whitelist/blacklist for tool access control |
| SSRF protection | IP blacklist + domain whitelist blocks internal network requests |
| Injection detection | 17 heuristic patterns prevent prompt/shell/SQL injection |
| Rate limiting | Token bucket per-tool, default 60 req/min |
| Audit logging | Structured JSON audit log with rotation + gzip |
AI Agent β mcp-slim-guard (security + compression) β agent-search-mcp
Be the first to review this server!
by Modelcontextprotocol Β· Developer Tools
Web content fetching and conversion for efficient LLM usage
by Toleno Β· Developer Tools
Toleno Network MCP Server β Manage your Toleno mining account with Claude AI using natural language.
by mcp-marketplace Β· Developer Tools
Create, build, and publish Python MCP servers to PyPI β conversationally.