Searching 105 engines at once mostly produces noise; naming the profile is what makes the result usable — a package search that looks at npm, PyPI and crates.io answers a different question than a general web search over the same words. The project reports 94.3% on the SimpleQA benchmark, which is a claim you can check against the comparison page it links.
A hosted MCP server for research. There is nothing to run: point your client at the URL and the assistant gains a question-answering path, a broad web search, a page-to-markdown extractor, and academic paper search. Search is fanned across many engines rather than routed to one.
- A factual question answered with its sources attached, rather than a link list to read yourself
- Web search across 105 engines, narrowed by one of 14 profiles — general, news, academic, programming, developer news, finance, code repositories, packages, security advisories, social, reference, images, video, or all engines at once
- Clean text or markdown extracted from any URL
- Academic search over arXiv, PubMed and Semantic Scholar
- Several searches issued in parallel when one query is not enough
- The same capabilities over plain HTTP at `https://contextwire.dev/api/ask`, `/api/search` and `/api/extract`, and through the `@contextwire/sdk` package
An API key from contextwire.dev; the free tier is 1,000 queries a month with no credit card. Nothing to install — add `https://contextwire.dev/mcp` as the server URL in your client config. The HTTP API takes the key as `Authorization: Bearer`. You can bring your own LLM key to spend fewer credits. MIT.
One command — npx -y mcp-remote https://contextwire.dev/mcp