The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.
MetaSearchMCP
Open-source metasearch backend, MCP server, and AI search API for LLM agents. Python FastAPI search gateway with Google search via SerpBase and Serper, multi-engine search…
Install
$
docker run --rm -p 8000:8000 --env-file .env metasearchmcpClaude Code
claude mcp add metasearchmcp -- docker run --rm -p 8000:8000 --env-file .env metasearchmcpCursor / Claude Desktop (mcp.json)
{
"mcpServers": {
"metasearchmcp": {
"command": "docker",
"args": [
"run",
"--rm",
"-p",
"8000:8000",
"--env-file",
".env",
"metasearchmcp"
]
}
}
}Install command taken from the project's README or registry entry. Always review code before running it.
README excerpt
Read full README on GitHub →MetaSearchMCP
Open-source metasearch backend for MCP, AI agents, and LLM workflows.
MetaSearchMCP aggregates results from multiple search providers, normalizes them into a stable JSON schema, and exposes both an HTTP API and an MCP server for agent tooling.
Positioning
- MCP-first metasearch backend
- Structured search API for AI pipelines
- Multi-provider search orchestration with deduplication and fallback
- Python FastAPI alternative to browser-first metasearch projects
Why It Exists
Most search aggregators are designed around browser UX: HTML pages, pagination, and interactive result cards. Agents and LLM workflows need a different contract: predictable JSON, stable field names, partial-failure tolerance, and provider-level execution metadata.
MetaSearchMCP is built for that machine-consumable workflow. The design is centered on search orchestration, normalized contracts, and MCP integration.
Core Features
- Concurrent multi-provider aggregation
- Unified result schema for web, academic, developer, and knowledge sources
- Provider-level timeout isolation and partial-failure handling
- Result deduplication across engines
- Provider selection by explicit names or semantic tags such as
web,academic,code, andgoogle - Final result caps for agent-friendly payload sizing
- HTTP API with OpenAPI docs
- MCP server over stdio for Claude Desktop, Cline, Continue, and similar clients
- Configurable provider allowlist via environment variables
Google Support
Google support now includes a direct scraper provider implemented inside this project.
The direct Google implementation uses browser-like requests, consent cookie handling, locale-aware query parameters, and resilient HTML result parsing. It is implemented locally in this repository.
Currently supported Google providers:
| Provider | Env var | Notes |
|---|---|---|
| Direct Google | ALLOW_UNSTABLE_PROVIDERS=true |
Primary path; HTML scraping, best effort, may be blocked from datacenter IPs |
| serpbase.dev | SERPBASE_API_KEY |
Pay-per-use; typically cheaper for low-volume usage |
| serper.dev | SERPER_API_KEY |
Includes a free tier, then pay-per-use |
Provider priority for /search/google is now google first, then google_serpbase, then google_serper.
Supported Providers
| Provider | Name | Method |
|---|---|---|
| Direct Google | google |
HTML scraping with browser-like request handling |
| SerpBase | google_serpbase |
Hosted Google SERP API |
| Serper | google_serper |
Hosted Google SERP API |
Web Search
| Provider | Name | Method |
|---|---|---|
| DuckDuckGo | duckduckgo |
HTML scraping |
| Bing | bing |
RSS feed |
| Yahoo | yahoo |
HTML scraping, best effort |
| Brave | brave |
Official Search API |
| You.com | youcom |
Official Search API |
| Mwmbl | mwmbl |
Public JSON API |
| Marginalia | marginalia |
Public JSON API, no key required |
| Ecosia | ecosia |
HTML scraping |
| Mojeek | mojeek |
HTML scraping |
| Startpage | startpage |
HTML scraping, best effort |
| Qwant | qwant |
Internal JSON API, best effort |
| Yandex | yandex |
HTML scraping, best effort |
| Baidu | baidu |
JSON endpoint, best effort |
| Seznam | seznam |
HTML scraping (Czech web), no key required |
| Naver | naver |
HTML scraping (Korean web), no key required |
| Ahmia | ahmia |
HTML scraping (Tor .onion services), no key required |
Similar plugs
Reverse engineer anything with agents, from app behavior down to native binaries.
Build Agentic workflows, RAG pipelines, with rich AI model and tool support on one collaborative workspace. Deploy on cloud, VPC, or self-hosted, so teams move from prototype…
User-friendly AI Interface (Supports Ollama, OpenAI API, ...)
A cross-platform desktop All-in-One assistant for Claude Code, Codex, OpenCode, OpenClaw, Grok Build & Hermes Agent. Only official website: ccswitch.io