Search, calculate and run-code are the standard gap-fillers. Adding a deploy step means the output of a session becomes a link you can send someone, rather than a code block they have to run themselves. That turns the toolkit from an assistant's scratchpad into something that produces artefacts. The separate arxiv and wikipedia searches are a small good decision too — both reward a targeted query far more than a general web search does.
302AI's toolkit for the capabilities a model does not have on its own: reaching the web, doing arithmetic reliably, running code, and reading files.
- Search — webSearch, wikipediaSearch and arxivSearch, plus webCrawl for a specific page
- Compute — calculator, and directRunCode for anything arithmetic cannot cover
- Files — parseFileToText
- Publishing — deployHtml and listHtmlProjects, so a generated page gets a URL
Npx. MIT licensed, and this is the vendor's own package. Documentation is available in English, Chinese and Japanese.
One command plus a key — npx -y @302ai/basic-mcp, then supply credentials