Lately I haven't been very happy with Kagi. A few searches have completely failed for me returning a blank page when both DuckDuckGo and Google had a response. The Kagi translate on/off switch made me think their services are unreliable. Then this happens and it feels like more that their word is unreliable. They expanded into too many domains and spread themselves too thin and that doesn't give me confidence.
My subscription is up for renewal in December and I think I actually have a decision to make. For now I'm very happy with the Kagi MCP for my research agent. But I'm definitely having a moment of "how reliant should I really be on this company?"
If you're mainly using the MCP, it could be worth making your own MCP that uses your own API keys from Brave, Mojeek & Marginalia (optionally also add Yep). For the most basic ranking, get it to group URLs that appear in more than one search engine and list those ones at the top. Your favorite AI will be able to vibe-build this all for you and give you an exe / binary you can run as an MCP that runs locally over stdio.
If you try it out, you'll either find a replacement that saves you money, or you'll at least have the makings of an escape plan if you need it later. You'll also have your own comparison data for where Kagi beats a generic metasearch, and if that difference is meaningful for you.
> They expanded into too many domains and spread themselves too thin and that doesn't give me confidence.
That's because imo there is no way to grow this bigger with a search service. Same for open source browsers, just look at Firefox instead of focusing on bugzilla and push for new web apis, they're trying very hard everything else than can be monetized