Open source
A2A-ready
MCP-compliant

The agentic server framework.

A single high-performance engine powering agents, tools, prompts, data plane connections, auth, cache, and telemetry.

Star on GitHub Agents quickstart →
curl -fsSL /install | bash

How it fits

Performant engine.
Unified outcome.

Same declarations, same artifact, same ops — not two stacks.

Build agents Build MCP

A2A

Agents

Build custom agents with your own models, and get ready endpoints to connect them to any agent harness.

MCP

Tools

Create tools that any agent harness can call, with built-in input validation and caching.

MCP

Resources

Ship resources that any harness can read and use for better context.

MCP

Prompts

Pack reusable conversation scaffolding, and let users use them with natural language.

Runtime

One binary, two surfaces

Your full agentic surface, all in one binary. No sidecars, no extra processes. No complexity.

Capabilities

Build an agentic server
from one declarative stack.

Only agents, tools, resources, prompts, or all. Same engine and deploy story.

Declarative A2A agents

Agent configs - models, tool access, validation.

Model providers

Bring your own models - OpenAI, Anthropic, Google, and more.

Two-tool discovery

Clients see only search and execute. Huge catalogs stay off the context window.

Full MCP surface

Tools, resources, and prompts — one spec-aligned server for clients and agents.

Built-in auth

Per-tool auth (api_key or custom) before every execute.

Data planes

Postgres, MySQL, Mongo, Redis, SQLite — pooling and health checks included.

Observe everything

OpenTelemetry and structured logs across search, execute, and agent paths.

Infinite scalability

Scale your agentic surface to any number of agents and tools, with a single binary.

Reality check

Fold the stack.
Keep the ambition.

Agents and capabilities usually ship as two stacks. Hyperterse is a single runtime for both. One build, one observability path.

Patchwork

When everything is “just one more service”

  • Two codebases, two deploys, two on-call stories for agents vs. clients.
  • Tool sprawl fills context; every new capability means another integration patch.
  • Auth, cache, and traces reimplemented — or skipped — per service.
  • Connection pools and health checks owned by whoever had time last sprint.
Hyperterse

When the surface is allowed to be whole

  • Agents, tools, prompts, resources — declared together, shipped as one artifact.
  • Discovery stays thin: clients find what they need without bloating every turn.
  • Same auth model, telemetry, and data adapters for autonomous and client traffic.
  • Pooling, shutdown, and production edges handled in the engine — not in your weekend.

Open Source

Built in the open.

Apache 2.0. Agents + MCP in one — star the repo.

Star on GitHub

or install now

curl -fsSL /install | bash

FAQ

Questions & answers.

What is Hyperterse?

Open-source agentic server framework: declare agents and an MCP server in one repo, one runtime — A2A agent HTTP, full MCP, search/execute on tools, plus auth, cache, and telemetry.

What does A2A compatibility mean here?

Agents follow the Agent2Agent protocol (standard HTTP/wire) for interoperability. Pre-2.5 agent configs are dropped — migrate via agent docs and schema.

Is Hyperterse free?

Yes — Apache 2.0, self-host free. Enterprise: enterprise@hyperterse.ai.

How do declaration files work in Hyperterse?

You declare database adapters, SQL-backed tools (typed inputs, optional auth and caching), and agents as config. Hyperterse discovers them from your project layout and compiles them — tool names follow your declarations, not boilerplate registration.

Which databases does Hyperterse support?

PostgreSQL, MySQL, MongoDB, Redis, SQLite (incl. libSQL/Turso). Multiple connections per instance; pooling and health checks included.

What is MCP?

Standard for tools, resources, and prompts for AI clients. Hyperterse exposes a full MCP server over Streamable HTTP (JSON-RPC 2.0); agents share the same backend.

Is my data secure with Hyperterse?

Clients never see connection strings or raw SQL — only typed MCP tools. Per-tool auth via api_key or custom plugins.

Does Hyperterse require writing code?

Mostly no — SQL and config in declarations. Optional TypeScript handlers for custom validation, transforms, or non-DB logic (sandboxed).

Is Hyperterse an ORM?

No — you write SQL in declarations. Hyperterse is about serving that as MCP/tools with production plumbing, not ORM or migrations.