- Published on
Show HN: CATT 0.2.1: Test agents/services against a Muse/Instinct-alike, in CI
- Authors
- Name
- DeepHorus
CATT 0.2.1, a new Consumer Agent Testing Tool, lets developers test how an assistant uses their service through scripted or model-driven agents against MCP servers, MCP apps, websites, or HTTP APIs. It checks what the service did, exposes evidence, and compares results across builds. It targets CI workflows where assistant-facing behavior must be verified repeatedly.
The tool supports Muse and Instinct as surrogate targets, with Muse as the default. Both use a shared connector testing contract covering routing, memory, payments, browser and HTTP tasks, and scheduling. CATT keeps the model driving a trial separate from the assistant being emulated.
CATT is explicitly a surrogate, not a certification. A passing test does not prove compatibility with Muse or any hosted assistant. Each Muse-specific behavior is recorded as documented, unofficial, assumed, or unknown. The project is not affiliated with or endorsed by Meta, Instinct, OpenAI, or Anthropic.
A local booking example runs under Instinct without a model or account. Users set emulation_target: instinct or pass --emulation-target instinct. Instinct defaults to plain-text replies and skips Muse guideline checks, while a local wallet fixture covers payment behavior.
CATT requires Python 3.12 or newer, uv, and Node.js 20 or newer with npm. CI tests Python 3.12 on Linux and 3.14 on macOS, and Node 25.9. Recorded release evidence covers macOS; Linux commands match CI but are not qualified.
Three quickstarts are provided. The connector lane and popular MCP server quickstarts run on loopback without an account, while the Coverage Cat sandbox quickstart calls a hosted sandbox. CATT uses the MIT license.
For teams building assistant-integrated services, CATT offers repeatable behavior checks before deployment. Its limits are clear: Linux support is uneven, and macOS Seatbelt-dependent features are unsupported or blocked on Linux. Adoption will depend on how well its surrogate tests track real assistant behavior.
