The declarative MCP framework.
The open-source framework for building MCP tool servers declaratively. Define tools once and get auth, caching, and observability built in. No glue code.
curl -fsSL /install | bash
Capabilities
Everything you need.
Nothing you don't.
From prototypes to multi-agent production systems — without changing your architecture.
Two-Tool Interface
Agents see exactly two MCP tools — search and execute. Define hundreds of tools behind them; the agent discovers what it needs dynamically.
Declarative Config
Define tools in declaration files. Each directory under app/tools/ becomes a tool the agent can search for and execute — no registration code.
Built-in Auth
Attach authentication to any tool with the built-in api_key plugin, or write your own. Auth runs before every execute call automatically.
TypeScript Scripts
Add handlers and transforms when declaration files alone aren't enough. Scripts run in a sandboxed runtime with fetch and console.
Caching
Enable result caching globally or per-tool with a TTL. Identical execute calls return cached results without hitting your database.
Multi-Database
Connect PostgreSQL, MySQL, MongoDB, and Redis. The framework manages pooling, health checks, and graceful shutdown.
Observability
Built-in OpenTelemetry tracing, metrics, and structured logging. Debug every search and execute call end-to-end.
Compile & Deploy
Build compiles your declaration files and scripts into a single artifact. Serve it anywhere — Docker, Kubernetes, bare metal, or any cloud.
How it works
Define. Build. Serve.
Every Hyperterse server exposes exactly two MCP tools, no matter how many you define behind it. Agents discover and run your tools through a single, consistent interface.
Define
Describe your tools, database connections, and authentication in declaration files. Add as many as you need — Hyperterse compiles them into a single artifact.
Search
The agent calls search with a natural-language query. Hyperterse ranks your tools by metadata — name, description, statement, inputs — and returns the best matches.
Execute
The agent picks a tool and calls execute with structured inputs. Hyperterse handles auth, validation, caching, and observability — then returns the result.
Why Hyperterse
Stop writing boilerplate.
Traditional MCP servers expose every tool individually. Hyperterse collapses them all behind search and execute — agents discover what they need, then run it.
- Every tool is a separate MCP endpoint the agent must know about
- Tool catalog bloats the agent's context window
- Write custom input validation and error handling
- Implement auth, caching, and observability yourself
- Manage database connections, pooling, and health checks manually
- Adding tools means updating the agent's integration
- Weeks of development and ongoing maintenance
- Agents see two tools: search and execute
- Tools discovered dynamically — no context window bloat
- Define each tool in declaratively
- Automatic input validation and type checking
- Auth, caching, and OpenTelemetry observability built in
- Framework handles pooling, health checks, and graceful shutdown
- Add 100 tools — agents still connect the same way
Open Source
Built in the open.
Hyperterse is free and open source under Apache 2.0. Two tools, unlimited possibilities. Star the repo, report issues, contribute features — the roadmap is shaped by the community.
Star on GitHubor install now
curl -fsSL /install | bash
FAQ