a2a-protocol-validator - Authentication for Agent Card

[rsoika]

We developed a new Enterprise system providing a A2A Agent card.
I found this new online validator: https://a2aprotocol.ai/a2a-protocol-validator
But my Agent Card is of course secured and not public as it describes enterprise internal operational processes.
How can I validate my agent card - an upload feature would be great for testing.

In general terms. Am I correct in assuming that an A2A system is usually a protected system within an enterprise IT landscape? That is, a user must always authenticate themselves against such a system before they can use it?

[lukehinds]

I’m not sure which language you’re using, but in Python, for example, the AgentCard is validated using the A2A-type models which has pydantic as its validation mechanism. Personally, I wouldn’t trust that validator - the whole site looks like it was vibecoded together.

If you're concerned about exposing too much of the agent’s capabilities to unauthorized entities through an unprotected AgentCard endpoint, but still want the card to be discoverable, you could use the Extended Agent Card instead. This allows you to protect more sensitive information—such as capabilities, skills, etc. - so that it's only visible to users who have been properly authorized.

https://a2aproject.github.io/A2A/specification/#91-fetching-authenticated-extended-agent-card

[rsoika]

Thanks for the answer. Yes I thinks this is fine for me. The public a2a-protocol-validator seems to be more a small tool for public services.
I implemented my server code in Java/Jakarta EE and I do also use a JSON wrapper object class so the validity of my outcome should be fine.
I have also seen that you can protect your A2A service with a bearer token authentication - which I already have implemented on my service endpoint. So my current understanding from this A2A and MCP stuff is:

1. A user is working with some LLM Chat/Client tool.
2. The tool is configured to use my A2A Server Endpoint
3. If the Chat/Client Tool decides to use my service, it will open a browser window and redirect the user to an Identity Provider specified in my AgentCard
4. When the authentication flow was successful, the Chat/Client Tool sends all requests to my endpoint with the received access token

So my own Rest service does not have to deal with the authentication process at all. It is similar to a normal protected web application - right?

[lukehinds]

Sorry for late reply, to be honest I don't think the spec has even approached auth flow yet, they just state the auth types which are borrowed from openAPI spec. It is something I am attempting soon and have been thinking about the same. Happy to chat more about it, but not to advise, more someone else puzzling through.

[kthota-g]

@rsoika, as @lukehinds suggested you can validate the agent card json with the types that are shipped part of the python sdk. Here is an example snippet for validating the agent card json.

import json

from pathlib import Path

from a2a.types import AgentCard
from pydantic import ValidationError


def validate_agent_card(file_path: str):
    """Validate the AgentCard JSON file against the Pydantic model."""
    print(f"--- Attempting to validate file: {file_path} ---")
    try:
        with Path(file_path).open('r') as f:
            data = json.load(f)

        # Create an instance of the AgentCard model for validation
        agent_card = AgentCard(**data)

        print("\n✅ Validation Successful!")
        print("Pydantic Model Object:")
        print(agent_card.model_dump_json(indent=4))

    except FileNotFoundError:
        print(f"\n❌ Error: The file '{file_path}' was not found.")
    except json.JSONDecodeError as e:
        print(f"\n❌ Invalid JSON format in '{file_path}': {e}")
    except ValidationError as e:
        # Catches Pydantic validation errors
        print("\n❌ Validation Failed!")
        print("Pydantic Validation Errors:")
        print(e)

if __name__ == "__main__":
    file_path = "agent_card.json"
    validate_agent_card(file_path)