fix(cli): resolve UnboundLocalError for json when serving A2A agents

[RUI-LONG]

I am following the official document, but the A2A server fails to expose any agent endpoint.

What I did

Following the quickstart, I started the API server with A2A enabled from an agent directory that contains an agent.json card:
adk api_server --a2a --port 8001 .

What I expected

The A2A agent to be registered and its endpoints (/a2a/<app_name> RPC route and /a2a/<app_name>/.well-known/agent-card.json) to be served, as described in the doc.

What actually happened

The server starts, but every A2A agent fails to register. The relevant log:

INFO  - fast_api.py:704 - Setting up A2A agent: check_prime_agent
ERROR - fast_api.py:739 - Failed to setup A2A agent check_prime_agent: cannot access local variable 'json' where it is not associated with a value

Problem:

Starting the API server with A2A enabled fails to register any A2A agent. Every
agent that ships an agent.json card is silently skipped, so its
/a2a/<app_name> RPC route and the corresponding
/.well-known/agent-card.json endpoint are never mounted.

Repro (a single agent directory containing agent.json):

$ adk api_server --a2a --port 8001 .
...
INFO  - fast_api.py:704 - Setting up A2A agent: check_prime_agent
ERROR - fast_api.py:739 - Failed to setup A2A agent check_prime_agent: cannot access local variable 'json' where it is not associated with a value

Root cause is a Python scoping issue in get_fast_api_app(). json is imported
at module level, but the function also imports it locally inside the
if gemini_enterprise_app_name: branch:

def get_fast_api_app(...):
    ...
    # A2A setup block (reached when --a2a is passed)
    with (p / "agent.json").open("r", encoding="utf-8") as f:
        data = json.load(f)          # (1) executed here
        agent_card = AgentCard(**data)
    ...
    if gemini_enterprise_app_name:
        ...
        import inspect
        import json                   # (2) makes `json` local to the WHOLE function

Because Python binds any name that is assigned anywhere in a function
(import counts as an assignment) as local to the entire function, the
module-level json is shadowed throughout get_fast_api_app(). At (1) the A2A
block runs json.load(f) before the local import json at (2) has executed, so
it dereferences an unbound local and raises:

cannot access local variable 'json' where it is not associated with a value

The except Exception guarding A2A setup swallows this into the
Failed to setup A2A agent ... log line, which is why the server still starts
but exposes no A2A endpoints — making the failure easy to miss.

Solution:

Remove the redundant local import json. It duplicates the module-level import
on line 19; deleting it stops json from being treated as a function-local and
lets json.load / json.dumps resolve to the module-level json as intended.

This is the minimal, single-line fix and keeps the imports in one place rather
than adding yet another local re-import to work around the shadowing.

     import inspect
-    import json

     from google.adk.agents import Agent

Testing Plan

Unit Tests:

• I have added or updated unit tests for my change.
• All unit tests pass locally.

Not added yet — happy to add a regression test under
tests/unittests/cli/ that calls get_fast_api_app(..., a2a=True) against a
temp agents dir with an agent.json and asserts the /a2a/<app> route is
mounted (it currently would not be). Guidance on the preferred fixture welcome.

Checklist

• I have read the CONTRIBUTING.md document.
• I have performed a self-review of my own code.
• I have commented my code, particularly in hard-to-understand areas.
• I have added tests that prove my fix is effective or that my feature works.
• New and existing unit tests pass locally with my changes.
• I have manually tested my changes end-to-end.
• Any dependent changes have been merged and published in downstream modules.

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.