Replies: 4 comments · 6 replies
-
|
Okay, I'll add some other search MCP options for users to choose from later. With exa, it's easy to get a $30 credit—$10 is given upon registration, and $20 is awarded after completing beginner tasks. Once used up, you can switch to another account and continue. |
Beta Was this translation helpful? Give feedback.
-
|
It all goes under the "don't track/spam me bro" banner. Can't be bothered signing up if there are "deep research" tool that can get similar level of quality (I can't stand Perplexity and how they screw up queries). |
Beta Was this translation helpful? Give feedback.
-
|
Some of these tools will need to be turned into an MCP sooner or later
If you want to promote Exa, then this would be top of the list https://github.com/zaidmukaddam/scira P.S. what is with Tavily and Linkup? Those are also what OpenHands support! https://github.com/MODSetter/SurfSense https://github.com/AnotiaWang/deep-research-web-ui P.S. Also what is with Serper!? |
Beta Was this translation helpful? Give feedback.
-
|
Still kind of a mind-blow that web search agents need prompting in annoying ways, like how they are not explorative thinkers, can't handle few-shot prompts, and won't do multi-part reasoning tasks. This explains why having multiple models or good, and why multi-agent system is more likely to scale in effectiveness. https://docs.perplexity.ai/guides/prompt-guide |
Beta Was this translation helpful? Give feedback.
-
|
This is a great suggestion. I'm currently researching subagents, and we could add a dedicated agent for searching. |
Beta Was this translation helpful? Give feedback.
-
|
2.12.9 added open-websearch mcp |
Beta Was this translation helpful? Give feedback.
-
|
Good defaults, will need to keep an eye on alternatives for Ecosia, Qwant, Startpage, Mojeek, SwissCow, Whoogle, and (self-host option) SearXNG. They might have bigger coverage for niche searches, but we never know. https://libretechtips.gitlab.io/detailed-tests-of-search-engines-google-startpage-bing-duckduckgo-metager-ecosia-swisscows-searx-qwant-yandex-and-mojeek/ For SearX https://github.com/ihor-sokoliuk/mcp-searxng https://github.com/SecretiveShell/MCP-searxng https://github.com/jae-jae/searxng-mul-mcp Addendum: according the following 2022 blogpost, Searx (still the best) > Mojeek, Qwant, Brave, MetaGer (basically the sites LibreTechTips graded badly) > Google, DDG, and Yandex (the usual players) |
Beta Was this translation helpful? Give feedback.
-
|
I think I have a clue why SearX is so good... they anonymize Google traffic, and then supplement with DDG and StartPage when big companies want to "memory-hole" niche items to save on server costs (hmm maybe this can be decentralized further but that is another project) https://docs.searxng.org/user/configured_engines.html#configured-engines We also need source coverage checks not just uptime checks https://searx.space https://uptime.searxng.org |
Beta Was this translation helpful? Give feedback.
-
|
In case people want a faster alternative https://github.com/neon-mmd/websurfx https://www.reddit.com/r/selfhosted/comments/134wm9k/websurfx_vs_searx_vs_searxng_comparison_of_the/ |
Beta Was this translation helpful? Give feedback.
-
|
Is Search based on this? (default is Addendum https://github.com/vishalkg/web-search?tab=readme-ov-file#-features |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
-
This happened when I let it use the default web search (not Exa)
Maybe there needs to be more options for web search, like Brave, DuckDuckGo, StartPage, MetaGer, SwissCow, or public SearX/SearXNG instances? Ideally they don't require API keys to function. (RIP Whoogle)
Beta Was this translation helpful? Give feedback.
All reactions