Karada

Not yet independently verified. The launch date rests on the maker’s own launch post of 23 September 2026. The site names no legal entity, the token-saving figures are the vendor’s own, and no independent report of the product was found. We will update this when it can be confirmed, and remove this note.

Karada turns an existing web API into a hosted Model Context Protocol server that AI agents can call, and was shown by its maker on 23 September 2026. It reads OpenAPI 3.0 or 3.1, Swagger 2.0 or Postman v2.1 collections, from a URL or an uploaded file, and generates a standalone Go server with typed structures and one tool definition per endpoint, served over streamable HTTP. The servers sit behind a gateway with isolated gateways per project, audit logging, scope management, OAuth 2.0 with PKCE, encrypted storage of the upstream API keys and rate limiting; optional plug-ins send failures to Sentry, usage to GA4 or PostHog and alerts to Slack. Pricing on its own page: Hobby free for five servers, Gateway Pro $19 a month, All-Access Pro $49 a month, and Enterprise with SSO, SAML and SLAs on request. The vendor’s figure of 40 to 60 per cent fewer tokens through its server-side agent is its own. The buyer’s question is the one every MCP gateway raises: one endpoint per tool means every API operation you publish becomes something an agent can call, so the scope settings matter more than the generator.