Exa Search

Exa is a neural search engine that actually understands what you mean, not just what you type. Unlike Google or traditional keyword search, it ranks pages by semantic similarity, which makes it absurdly good at "find me things like this" queries. I've used it to map out entire research domains by feeding it one good paper and asking for similar content. The MCP server wraps Exa's API cleanly, so Claude can run these searches mid-conversation without you copy-pasting URLs. What it's genuinely good for: research discovery, competitive analysis, finding obscure technical content that keyword search misses. If you're building a literature review, tracking what competitors are publishing, or hunting down niche blog posts on a specific architecture pattern, Exa will surface things you'd never find otherwise. The "similar to this URL" feature is the killer workflow. I'll often start with one authoritative source, then ask Claude to find 10 similar pages and summarise the common themes. It's faster than manual trawling and catches stuff I'd have missed. Quirks: Exa is a paid API, so you'll burn through credits if you're not deliberate about queries. The free tier exists but won't last long if you're doing serious research. Also, because it's neural, results can feel opaque. Sometimes it surfaces a page and you have to squint to see why it matched. That's the trade-off for semantic understanding. If you need exact keyword matches or want to filter by date ranges with precision, stick to traditional search. Who shouldn't bother: if you're just looking up documentation or doing basic web lookups, this is overkill. Use a simpler search MCP or just ask Claude to guess. Exa shines when the problem is discovery, not retrieval. If you don't have an Exa API key and aren't willing to pay for one, skip it. The server is useless without the key, and there's no localhost fallback.