MattMessinger1/agentic_refund_guardrail
Safe refund policy layer for AI agents — Python + TypeScript. Same behavior, shared tests.
Platform-specific configuration:
{
"mcpServers": {
"agentic_refund_guardrail": {
"command": "npx",
"args": [
"-y",
"agentic_refund_guardrail"
]
}
}
}Add the config above to .claude/settings.json under the mcpServers key.
[](https://github.com/MattMessinger1/agentic_refund_guardrail/actions/workflows/ci.yml) [](LICENSE) [](pyproject.toml) [](packages/refund-guard-ts/package.json)
Start here: Step-by-step guide · Integration guide · Contributing · Report an issue
A small library that turns one real order into one safe refund function for your AI agent — so the agent can only refund what your policy allows (window, amount cap, remaining balance).
> New here? Read docs/STEP_BY_STEP.md first, then come back for details.
---
| Question | Answer | |----------|--------| | Is this a hosted API or SaaS? | No. It is a package you install (pip / npm) in your server code. | | Does it run on my phone? | Not inside the app. Your mobile app calls your backend; the backend runs this library. | | Do I need Python *and* TypeScript? | No. Pick one — whatever your backend uses. | | What does it actually do? | Wraps your existing refund call with policy checks before money moves. | | How is this different from agent guardrail products (Veto, PolicyLayer, Kvlar, etc.)? | Those answer: *should this tool run at all?* refund-guard answers: *for this order and this amount, does our business policy allow it?* Use both if you want. | | Do I have to use a YAML file? | No. Pass a plain object: Refunds({ skus: { my_sku: { refund_window_days: 14 } } }). YAML is available when you have many SKUs or want non-engineers to ed
Loading reviews...