You are a senior infrastructure engineer and open-source architect.

I want to build an open-source Python package called `agentgateway`.

Vision
------
In the future, applications will have two interfaces:

1. Human interface (UI)
2. Agent interface (machine-native)

Instead of AI agents interacting with websites through HTML and clicking buttons, applications should expose a standardized agent layer that lets agents discover and execute capabilities safely.

The goal is to make existing backends agent-ready with one line of code.

Example:

from agentgateway import enable_agents

app = FastAPI()

enable_agents(app)

Core requirements
-----------------
1. Zero-friction integration.
2. Support FastAPI first.
3. No changes required to existing endpoints.
4. Automatically scan routes and Pydantic models.
5. Generate a machine-readable manifest.

Expose:

GET /.well-known/agent.json

Example output:

{
  "version": "1.0",
  "tools": [
    {
      "name": "add_to_cart",
      "description": "Add a product to the cart",
      "input_schema": {...}
    }
  ]
}

Architecture
------------
agentgateway/
    core.py
    scanner.py
    schema.py
    registry.py
    server.py
    decorators.py
    cli.py
    auth.py

Features for MVP
----------------
- enable_agents(app)
- Route scanning
- Tool generation from FastAPI routes
- Pydantic schema extraction
- Agent manifest endpoint
- @agent_tool decorator for overriding metadata
- agentctl CLI

Example:

@agent_tool(
    description="Create an order",
    approval_required=True
)
def create_order(...):
    ...

Future features (not MVP)
-------------------------
- Permissions and scopes
- Human approval flows
- Audit logs
- OpenAI/Anthropic MCP compatibility
- OAuth
- Memory
- Multi-agent orchestration
- Cloud platform

Important:
-----------
Do not overengineer.
Keep the MVP extremely simple and elegant.

Your task:

1. Critique the idea technically.
2. Identify existing standards and libraries that overlap with it.
3. Propose improvements.
4. Design the package structure.
5. Define the agent.json schema.
6. Suggest the best developer experience.
7. Generate a step-by-step implementation roadmap.
8. Build the MVP incrementally with production-quality code and tests.
9. Challenge assumptions instead of blindly agreeing.
10. Think like the creator of FastAPI or Terraform, prioritizing simplicity and adoption.