I’m Samrith, creator of Hyperterse.

Today I’m launching Hyperterse 2.0, a schema-first framework for building MCP servers directly on top of your existing production databases.

If you're building AI agents in production, you’ve probably run into agents needing access to structured, reliable data but wiring your business logic to MCP tools is tedious. Most teams end up writing fragile glue code. Or worse, giving agents unsafe, overbroad access.

There isn’t a clean, principled way to expose just the right data surface to agents.

Hyperterse lets you define a schema over your data and automatically exposes secure, typed MCP tools for AI agents.

Think of it as: Your business data → controlled, agent-ready interface.

Some key properties include a schema-first access layer, typed MCP tool generation, works with existing Postgres, MySQL, MongoDB, Redis databases, fine-grained exposure of queries, built for production agent workloads.

v2.0 focuses heavily on MCP with first-class MCP server support, cleaner schema ergonomics, better type safety, faster tool surfaces.

All of this, with only two tools - search & execute - reducing token usage drastically.

Hyperterse is useful if you are building AI agents/copilots, adding LLM features to existing SaaS, trying to safely expose internal data to agents or are just tired of bespoke MCP glue layers.

I’d love feedback, especially from folks running agents in production.

GitHub: https://github.com/hyperterse/hyperterse