🎉 Full MCP Spec + SQLite

This release completes MCP specification compliance and adds SQLite as a first-class connector. Hyperterse servers can now expose tools, resources, and prompts in one place — the three pillars of the MCP protocol. Add SQLite (local or Turso/libSQL) for lightweight backends, side‑project demos, or edge-deployed AI apps.

What you can build now: AI agents that read release notes, summarize incidents, or query SQLite — with prompt templates and static context discovered automatically. Zero extra setup beyond config.

Backward compatible — no migration required. Existing configs work as-is; add app/prompts/ and app/resources/ when you're ready.

✨ Features

MCP Resources & Prompts (Full Spec Compliance)

Hyperterse now implements the complete MCP resource and prompt surface. AI clients can discover and consume static context alongside callable tools.

• Resources — Expose read-only content via resources/list, resources/read, and resources/templates/list. Support concrete resources (fixed URI) and resource templates (parameterized uri_template with {{ id }} placeholders). Content can be inline (text, text_template) or file-backed (file, file_template).
• Prompts — Define reusable prompt templates in app/prompts/**/*.terse with argument interpolation, completion hints, and multi-message scaffolding. Exposed through prompts/list and prompts/get; clients get rendered messages with {{ argument }} replaced.
• Completion API — completion/complete supports both prompts and resource templates, so clients get typeahead and validation for template arguments when you declare completion values.
• Subscriptions & notifications — resources/subscribe / resources/unsubscribe; notifications/resources/updated, notifications/prompts/list_changed for live updates after model reload.
• Discovery conventions — Config in app/prompts/ and app/resources/; declarative YAML with MIME types, descriptions, and optional argument schemas.

SQLite Connector

A new database connector for SQLite: local files, in-memory, or remote libSQL/Turso.

• Local — file:./app.db, :memory:, or absolute paths for embedded and dev workloads.
• Remote (libSQL / Turso) — libsql:// or https:// for hosted Turso; http:// for self-hosted libSQL. Auth via authToken, auth_token, or jwt query params; TLS control via ?tls=0|1.
• Adapter config — Uses app/adapters/*.terse like PostgreSQL and MySQL; same patterns for connection strings, env substitution, and options.
• Execution — Parameterized queries, transaction handling, OpenTelemetry tracing, and observability metrics. Compatible with the existing tool executor and framework.

Full Changelog: v2.2.1...v2.3.0

Bug fixes

• Init command directory path — Fixed directory references from .agent to .agents when scaffolding for Hyperterse integration with AI agents

🔧 Improvements

• Agent skills documentation — Updated the documentation link in SKILL.md to point to the correct resource
• Init output messaging — Revised the output message for running the Hyperterse command after scaffolding

Full Changelog: v2.2.0...v2.2.1

✨ Features

• Smarter MCP search relevance ranking — Search scoring now adapts to the metadata each tool actually has, improving hit quality across SQL-backed and handler-only tools

🔧 Improvements

• Handler-only tool relevance — Handler-only tools no longer receive statement-based relevance from potential synthetic placeholder statements
• Conversational query matching — Token scoring now applies a soft miss penalty so intent-heavy natural language prompts rank more accurately
• Search ranking test coverage — Added targeted tests for handler-only scoring, conversational queries, and active-weight normalization

Bug fixes

• Search payload shape update — MCP search results no longer include statement; clients should rely on name, description, relevance_score, and inputs

Full Changelog: v2.1.0...v2.2.0

✨ Features

• Init command folder support — hyperterse init now accepts an optional folder name as argument; scaffolds into <folder>/.hyperterse by default
• Script-backed tool scaffolding — Init command generates script-backed tool structure instead of adapter-based config; includes greeting handler example
• Agent skills documentation — New agent skills directory with documentation for Hyperterse integration with AI agents

🔧 Improvements

• Init output path logic — Output file defaults to <folder>/.hyperterse when folder is specified; prompts for folder name when not provided
• Documentation updates — Revised quickstart and CLI reference to reflect script-backed tools and new project structure

Full Changelog: v2.0.0...v2.1.0

🎉 Hyperterse v2.0.0 — Tool-First MCP Framework

Hyperterse v2 is a breaking redesign. The framework shifts from a single-file query gateway to a compiled, filesystem-based MCP framework. Define tools in declarative config, add optional TypeScript handlers, and serve them as a standards-compliant MCP server over Streamable HTTP.

✨ Features

• Filesystem-based tool discovery — Each directory under app/tools/ maps to one MCP tool; no manual registration
• Declarative tool config — Root config + app/adapters/*.terse + app/tools/*/config.terse for clear separation of concerns
• Embedded scripting — TypeScript handlers and input/output transforms for logic that config alone cannot express; scripts bundled at compile time
• Per-tool authentication — Pluggable auth per tool: allow_all, api_key, or custom plugins
• Build and serve pipeline — hyperterse build compiles config and scripts into a deployable artifact; hyperterse serve runs from the artifact
• Search tool — Built-in MCP search tool with configurable result limit and statement support for tool discovery
• Graceful shutdown — BaseContext-based shutdown for in-flight handlers to complete before exit

🔧 Improvements

• MCP-only transport — Tools exposed exclusively via MCP Streamable HTTP (/mcp); /query/{name} removed
• Per-tool caching — Global cache config plus per-tool override for TTL and behavior
• OpenTelemetry observability — Distributed tracing, metrics, and structured logging out of the box
• CLI commands — start (with optional --watch), build, serve, validate, init; run replaced by start
• Project convention — .hyperterse root config; hyperterse start for development

⚠️ Breaking Changes

See the v1 to v2 migration guide for step-by-step migration instructions.

Full Changelog: v1.4.0...v2.0.0

This release turns Hyperterse into a very slim MCP which takes fewer tokens. It only exposes two tools - search & execute. This drastically reduces tool surface and makes it easier to use in smaller context windows.

✨ Features

• Search tool functionality — Added search tool support with configurable result limits for safer, more controllable query responses
• Condensed tool execution — Tools can now only be executed through a unified execute tool by passing their named identified and the required inputs

This is a release candidate of Hyperterse. We welcome feedback and contributions!
Full Changelog: v2.0.0-rc.0...v2.0.0-rc.1

🚨 Hyperterse v2 — Tool-First Rewrite (Breaking)

Hyperterse v2 is a major architectural release that moves the platform from a route/query-centric model to a tool-centric runtime.
This unlocks a cleaner execution model for APIs + MCP, but it also introduces intentional breaking changes for v1 users.

Important: This is a release candidate with large migration impact. If you’re on v1.x, plan and test migration before production rollout.

Why this release matters

• Hyperterse now treats tools as the primary unit of execution.
• Runtime behavior and configuration conventions have been aligned around this model.
• CLI and validation workflows were updated to support the new architecture end-to-end.
• Documentation was restructured to reflect the new mental model and usage patterns.

⚠️ Breaking changes

• Architecture change: route/query-based flows were refactored into a tool-based architecture.
• Config convention updates: project conventions now use .hyperterse-style workflows.
• CLI behavior updates: command surface and runtime expectations were adjusted to match the new model.
• Legacy docs/references removed: deprecated references and outdated docs were cleaned up.

✨ What’s new in v2.0.0-rc.0

• Added build and serve commands to improve operational workflows.
• Improved execution-mode validation for tools.
• Strengthened runtime error handling and diagnostics.
• Enhanced validate command behavior.
• Updated MCP integration and schema support.
• Distribution manifests updated for the v2 RC line.

Migration at a Glance (v1.x → v2)

• Review updated docs and README for the new tool-based model.
• Migrate route/query-style definitions to tool-style definitions.
• Update project/config conventions to current .hyperterse expectations.
• Run validation and regression checks for API + MCP behavior.
• Test release workflows if you rely on prerelease tagging/version automation.

This is a beta release of Hyperterse. We welcome feedback and contributions!
v2 docs: https://hyperterse.mintlify.app
Full Changelog: v1.4.0...v2.0.0-rc.0

✨ Features

• Query execution caching — Added executor-level query caching to reduce repeated work and improve runtime performance
• Config validation command — Added a dedicated CLI command to validate .terse configuration files before runtime

🔧 Improvements

• CLI error handling and logging — Improved command-level error handling and log output for clearer diagnostics

Full Changelog: v1.3.0...v1.4.0

✨ Features

• MongoDB connector — New database connector for MongoDB, including BSON-to-JSON parsing and support for MongoDB connection strings and queries

🐛 Bug Fixes

• MySQL connection strings — Fixed URL-to-DSN conversion for MySQL connection strings
• MongoDB JSON responses — Simplified and corrected JSON serialization of MongoDB query results

🔧 Improvements

• .env loading — Environment files are now loaded earlier and more reliably during startup

⚠️ Schema Changes

• Configuration schema updated to support mongodb as a connector type in adapters

Full Changelog: v1.2.1...v1.3.0

🐛 Bug Fixes

• NPM distribution — Fixed npm-installed CLI not forwarding command-line arguments to the binary; commands will now work correctly when installed via NPM.

This completely went under the radar, and should have been addressed from v1 itself. Apologies from the team!

Full Changelog: v1.2.0...v1.2.1