---
title: Choosing an integration path
description: Four integration layers, from raw HTTP to MCP, over one shared contract.
---

# Choosing an integration path

Four layers, from universal to most convenient. Every layer sits on the same HTTP
contract; pick the highest one your runtime supports.

| Your runtime | Use | Quickstart |
|---|---|---|
| Anything that speaks HTTP | The OpenAPI contract directly | [HTTP quickstart](http-quickstart.md) |
| Python agent or relying party | `yanez-agent-authorization` (import `yanez_authz`) | [examples/python/](https://github.com/yanez-compliance/yanez-agent-authorization/tree/main/examples/python) |
| Node agent or relying party | `@yanez.ai/agent-authorization` | [examples/typescript/](https://github.com/yanez-compliance/yanez-agent-authorization/tree/main/examples/typescript) |
| Shell-capable coding agent | `yanez-authz` CLI + the `yanez-authorize` skill | [examples/skill-cli/](https://github.com/yanez-compliance/yanez-agent-authorization/tree/main/examples/skill-cli) |
| MCP-capable host | `yanez-authz-mcp` (stdio) + the skill | [examples/mcp/](https://github.com/yanez-compliance/yanez-agent-authorization/tree/main/examples/mcp) |

MCP is **not required**. It is the preferred adapter when the host already supports it,
because it gives the model discoverable typed tools and keeps the credential in a
process the model never sees. A skill by itself is not a security boundary and cannot
make authenticated calls; pair it with the CLI or the MCP server.

Three parties, three responsibilities:

1. **The agent** creates an authorization request with exact terms and polls for the
   decision. Its `yak_` key can ask, not act.
2. **The user** approves or rejects in the YID app, gated on a fresh biometric scan.
3. **The action executor** (relying party) verifies the signed receipt against the
   proposed action and consumes it when single-use — [action enforcement](action-enforcement.md).

Credential rules, everywhere: the `yak_` key comes from configuration
(`YANEZ_AGENT_API_KEY` or a secret manager), never from model prompts, tool arguments,
command-line flags, or logs. Never hand it to a sub-agent.

## Install

The Python SDK is on PyPI as a pre-release. Pass `--pre`, because pip prefers stable
releases and skips pre-releases unless you ask for them:

```sh
pip install --pre yanez-agent-authorization
```

For a reproducible install, pin the exact build instead, for example
`pip install yanez-agent-authorization==0.1.0b4`. The
[release history](https://pypi.org/project/yanez-agent-authorization/#history) lists
every published version.

The CLI and the MCP server are not published yet. Until they are, install them from
a checkout. See Development in the
[repository README](https://github.com/yanez-compliance/yanez-agent-authorization#development).

| Path | Install | Status |
|---|---|---|
| Python SDK | `pip install --pre yanez-agent-authorization` | PyPI, pre-release |
| CLI | `pip install yanez-authz-cli` (installs `yanez-authz`) | Not published; install from a checkout |
| MCP server | `pip install yanez-authz-mcp` | Not published; install from a checkout |
| TypeScript SDK | `npm install @yanez.ai/agent-authorization@beta` | npm, pre-release |