feat: Add --agent_engine_config_file option to adk deploy agent_engine

PiperOrigin-RevId: 800171071

CHANGELOG.md

@@ -1,5 +1,53 @@
 # Changelog
 
+## 1.12.0 (2025-08-21)
+
+### Features
+
+**[Agent Config]** 🌟 **NEW FEATURE**: Support using config file (YAML) to author agents in addition to python code. See the [documentation](https://google.github.io/adk-docs/agents/config/) for details.
+* [Agent Config] Support deploying config agent to Agent Engine in CLI ([b3b7003](https://github.com/google/adk-python/commit/b3b70035c432670a5f0b5cdd1e9467f43b80495c))
+* [Tools] Add a dedicated Bigtable toolset to provide an easier, integrated way to interact
+with Bigtable for building AI Agent applications(experimental feature) ([a953807](https://github.com/google/adk-python/commit/a953807cce341425ba23e3f0a85eae58d6b0630f))
+* [Tools] Support custom tool_name_prefix in auto-generated GoogleApiToolset ([a2832d5](https://github.com/google/adk-python/commit/a2832d5ac7ba5264ee91f6d5a6a0058cfe4c9e8a)) See [oauth_calendar_agent](https://github.com/google/adk-python/tree/main/contributing/samples/oauth_calendar_agent) as an example.
+* [CLI] Add `build_image` option for `adk deploy cloud_run` CLI ([c843503](https://github.com/google/adk-python/commit/c84350345af0ea6a232e0818b20c4262b228b103))
+* [Services] Add setdefault method to the ADK State object ([77ed1f5](https://github.com/google/adk-python/commit/77ed1f5f15ed3f009547ed0e20f86d949de12ec2))
+
+
+### Bug Fixes
+
+* Lazy load VertexAiCodeExecutor and ContainerCodeExecutor ([018db79](https://github.com/google/adk-python/commit/018db79d1354f93b8328abb8416f63070b25f9f1))
+* Fix the path for agent card in A2A demo ([fa64545](https://github.com/google/adk-python/commit/fa64545a9de216312a69f93126cfd37f1016c14b))
+* Fix the path for agent card in A2A demo ([a117cf0](https://github.com/google/adk-python/commit/a117cf0af335c5e316ae9d61336a433052316462))
+* litellm-test due to breaking change in dep library of extension extra ([004a0a0](https://github.com/google/adk-python/commit/004a0a0f2d9a4f7ae6bff42a7cad96c11a99acaf))
+* Using base event's invocation id when merge multiple function response event ([279e4fe](https://github.com/google/adk-python/commit/279e4fedd0b1c0d1499c0f9a4454357af7da490e))
+* Avoid crash when there is no candidates_token_count, which is Optional ([22f34e9](https://github.com/google/adk-python/commit/22f34e9d2c552fbcfa15a672ef6ff0c36fa32619))
+* Fix the packaging version comparison logic in adk cli ([a2b7909](https://github.com/google/adk-python/commit/a2b7909fc36e7786a721f28e2bf75a1e86ad230d))
+* Add Spanner admin scope to Spanner tool default Oauth scopes ([b66054d](https://github.com/google/adk-python/commit/b66054dd0d8c5b3d6f6ad58ac1fbd8128d1da614))
+* Fixes SequentialAgent.config_type type hint ([8a9a271](https://github.com/google/adk-python/commit/8a9a271141678996c9b84b8c55d4b539d011391c))
+* Fixes the host in the ansi bracket of adk web ([cd357bf](https://github.com/google/adk-python/commit/cd357bf5aeb01f1a6ae2a72349a73700ca9f1ed2))
+* Add spanner tool name prefix ([a27927d](https://github.com/google/adk-python/commit/a27927dc8197c391c80acb8b2c23d610fba2f887))
+
+### Improvements
+
+* Support `ADK_SUPPRESS_EXPERIMENTAL_FEATURE_WARNINGS` as environment variable to suppress experimental warnings ([4afc9b2](https://github.com/google/adk-python/commit/4afc9b2f33d63381583cea328f97c02213611529))
+* Uses pydantic `Field` for Agent configs, so that the generated AgentConfig.json json schema can carry field description ([5b999ed](https://github.com/google/adk-python/commit/5b999ed6fd23a0fc1da56ccff4c09621f433846b))
+* Update `openai` dependency version, based on correct OPENAI release ([bb8ebd1](https://github.com/google/adk-python/commit/bb8ebd15f90768b518cd0e21a59b269e30d6d944))
+* Add the missing license header for core_callback_config init file ([f8fd6a4](https://github.com/google/adk-python/commit/f8fd6a4f09ab520b8ecdbd8f9fe48228dbff7ebe))
+* Creates yaml_utils.py in utils to allow adk dump yaml in the same style ([1fd58cb](https://github.com/google/adk-python/commit/1fd58cb3633992cd88fa7e09ca6eda0f9b34236f))
+* Return explict None type for DELETE endpoints ([f03f167](https://github.com/google/adk-python/commit/f03f1677790c0a9e59b6ba6f46010d0b7b64be50))
+* Add _config suffix to all yaml-based agent examples ([43f302c](https://github.com/google/adk-python/commit/43f302ce1ab53077ee8f1486d5294540678921e6))
+* Rename run related method and request to align with the conventions ([ecaa7b4](https://github.com/google/adk-python/commit/ecaa7b4c9847b478c7cdc37185b1525f733bb403))
+* Update models in samples/ folder to be gemini 2.0+ ([6c217ba](https://github.com/google/adk-python/commit/6c217bad828edf62b41ec06b168f8a6cb7ece2ed))
+* Remove the "one commit" requirement from the contributing guide ([c32cb6e](https://github.com/google/adk-python/commit/c32cb6eef9ce320ea5a1f3845fc57b83762c237e))
+* Bump version to 1.11.0 ([8005270](https://github.com/google/adk-python/commit/80052700f6cee947322080ae6c415d3a428b6c91))
+
+### Documentation
+
+* Add contributing bigtable sample ([fef5318](https://github.com/google/adk-python/commit/fef5318a22f3dcaadb7ecb858725eb61a0350140))
+* Fix core_callback example ([ba6e85e](https://github.com/google/adk-python/commit/ba6e85eb3fb06f58ce9077574eac193298e18bea))
+* Adds a minimal sample to demo how to use Agent Config to create a multi-agent setup ([1328e6e](https://github.com/google/adk-python/commit/1328e6ef62e9e6260048c0078579edb85a0440bc))
+
+
 ## [1.11.0](https://github.com/google/adk-python/compare/v1.10.0...v1.11.0) (2025-08-14)
 
 ### Features

README.md

@@ -27,6 +27,11 @@ Agent Development Kit (ADK) is a flexible and modular framework for developing a
 
 ---
 
+## ✨ What's new
+
+- **Agent Config**: Build agents without code. Check out the
+  [Agent Config](https://google.github.io/adk-docs/agents/config/) feature.
+
 ## ✨ Key Features
 
 - **Rich Tool Ecosystem**: Utilize pre-built tools, custom functions,

autoformat.sh

@@ -52,16 +52,16 @@ echo '---------------------------------------'
 echo '|  Auto-formatting src/...'
 echo '---------------------------------------'
 
-find -L src/ -type f -name "*.py" -exec pyink --config pyproject.toml {} +
+find -L src/ -not -path "*/.*" -type f -name "*.py" -exec pyink --config pyproject.toml {} +
 
 echo '---------------------------------------'
 echo '|  Auto-formatting tests/...'
 echo '---------------------------------------'
 
-find -L tests/ -type f -name "*.py" -exec pyink --config pyproject.toml {} +
+find -L tests/ -not -path "*/.*" -type f -name "*.py" -exec pyink --config pyproject.toml {} +
 
 echo '---------------------------------------'
 echo '|  Auto-formatting contributing/...'
 echo '---------------------------------------'
 
-find -L contributing/ -type f -name "*.py" -exec pyink --config pyproject.toml {} +
+find -L contributing/ -not -path "*/.*" -type f -name "*.py" -exec pyink --config pyproject.toml {} +

contributing/samples/adk_pr_triaging_agent/agent.py

@@ -32,6 +32,7 @@ LABEL_TO_OWNER = {
     "documentation": "polong-lin",
     "services": "DeanChensj",
     "tools": "seanzhou1023",
+    "mcp": "seanzhou1023",
     "eval": "ankursharmas",
     "live": "hangfei",
     "models": "genquan9",
@@ -267,6 +268,7 @@ root_agent = Agent(
       - If it's about model support(non-Gemini, like Litellm, Ollama, OpenAI models), label it with "models".
       - If it's about tracing, label it with "tracing".
       - If it's agent orchestration, agent definition, label it with "core".
+      - If it's about Model Context Protocol (e.g. MCP tool, MCP toolset, MCP session management etc.), label it with "mcp".
       - If you can't find a appropriate labels for the PR, follow the previous instruction that starts with "IMPORTANT:".
 
       Here is the contribution guidelines:

contributing/samples/adk_triaging_agent/agent.py

@@ -31,6 +31,7 @@ LABEL_TO_OWNER = {
     "documentation": "polong-lin",
     "services": "DeanChensj",
     "question": "",
+    "mcp": "seanzhou1023",
     "tools": "seanzhou1023",
     "eval": "ankursharmas",
     "live": "hangfei",
@@ -182,6 +183,7 @@ root_agent = Agent(
       - If it's about tracing, label it with "tracing".
       - If it's agent orchestration, agent definition, label it with "core".
       - If it's about agent engine, label it with "agent engine".
+      - If it's about Model Context Protocol (e.g. MCP tool, MCP toolset, MCP session management etc.), label it with "mcp".
       - If you can't find a appropriate labels for the issue, follow the previous instruction that starts with "IMPORTANT:".
 
       Call the `add_label_and_owner_to_issue` tool to label the issue, which will also assign the issue to the owner of the label.

contributing/samples/bigquery/README.md

@@ -25,6 +25,16 @@ distributed via the `google.adk.tools.bigquery` module. These tools include:
 
   Runs a SQL query in BigQuery.
 
+1. `ask_data_insights`
+
+  Natural language-in, natural language-out tool that answers questions
+  about structured data in BigQuery. Provides a one-stop solution for generating
+  insights from data.
+
+  **Note**: This tool requires additional setup in your project. Please refer to
+  the official [Conversational Analytics API documentation](https://cloud.google.com/gemini/docs/conversational-analytics-api/overview)
+  for instructions.
+
 ## How to use
 
 Set up environment variables in your `.env` file for using

contributing/samples/live_bidi_streaming_single_agent/agent.py

@@ -65,8 +65,8 @@ async def check_prime(nums: list[int]) -> str:
 
 
 root_agent = Agent(
-    model='gemini-2.0-flash-live-preview-04-09',  # for Vertex project
-    # model='gemini-2.0-flash-live-001',  # for AI studio key
+    # model='gemini-2.0-flash-live-preview-04-09',  # for Vertex project
+    model='gemini-2.0-flash-live-001',  # for AI studio key
     name='hello_world_agent',
     description=(
         'hello world agent that can roll a dice of 8 sides and check prime'

contributing/samples/multi_agent_llm_config/README.md

@@ -1,3 +1,3 @@
 # Config-based Agent Sample - LLM multi-agent
 
-http://google3/third_party/py/google/adk/open_source_workspace/contributing/samples/hello_world_ma/
+From contributing/samples/hello_world_ma/

contributing/samples/spanner/README.md

@@ -23,7 +23,7 @@ distributed via the `google.adk.tools.spanner` module. These tools include:
 
 1. `get_table_schema`
 
-  Fetches Spanner database table schema.
+  Fetches Spanner database table schema and metadata information.
 
 1. `execute_sql`
 

contributing/samples/tool_human_in_the_loop_config/README.md

@@ -1,3 +1,3 @@
 # Config-based Agent Sample - Human-In-The-Loop
 
-http://google3/third_party/py/google/adk/open_source_workspace/contributing/samples/human_in_loop/
+From contributing/samples/human_in_loop/

contributing/samples/tool_mcp_stdio_notion_config/README.md

@@ -0,0 +1,48 @@
+# Config-based Agent Sample - MCP Toolset with Notion MCP Server
+
+This sample demonstrates how to configure an ADK agent to use the Notion MCP server for interacting with Notion pages and databases.
+
+## Setup Instructions
+
+### 1. Create a Notion Integration
+
+1. Go to [Notion Integrations](https://www.notion.so/my-integrations)
+2. Click "New integration"
+3. Give it a name and select your workspace
+4. Copy the "Internal Integration Secret" (starts with `ntn_`)
+
+For detailed setup instructions, see the [Notion MCP Server documentation](https://www.npmjs.com/package/@notionhq/notion-mcp-server).
+
+### 2. Configure the Agent
+
+Replace `<your_notion_token>` in `root_agent.yaml` with your actual Notion integration token:
+
+```yaml
+env:
+  OPENAPI_MCP_HEADERS: '{"Authorization": "Bearer secret_your_actual_token_here", "Notion-Version": "2022-06-28"}'
+```
+
+### 3. Grant Integration Access
+
+**Important**: After creating the integration, you must grant it access to specific pages and databases:
+
+1. Go to `Access` tab in [Notion Integrations](https://www.notion.so/my-integrations) page
+2. Click "Edit access"
+3. Add pages or databases as needed
+
+### 4. Run the Agent
+
+Use the `adk web` to run the agent and interact with your Notion workspace.
+
+## Example Queries
+
+- "What can you do for me?"
+- "Search for 'project' in my pages"
+- "Create a new page called 'Meeting Notes'"
+- "List all my databases"
+
+## Troubleshooting
+
+- If you get "Unauthorized" errors, check that your token is correct
+- If you get "Object not found" errors, ensure you've granted the integration access to the specific pages/databases
+- Make sure the Notion API version in the headers matches what the MCP server expects

contributing/samples/tool_mcp_stdio_notion_config/root_agent.yaml

@@ -13,4 +13,4 @@ tools:
       - "-y"
       - "@notionhq/notion-mcp-server"
       env:
-        OPENAPI_MCP_HEADERS: "{'Authorization': 'Bearer fake_notion_api_key', 'Notion-Version': '2022-06-28'}"
+        OPENAPI_MCP_HEADERS: '{"Authorization": "Bearer <your_notion_token>", "Notion-Version": "2022-06-28"}'

pyproject.toml

@@ -25,37 +25,38 @@ classifiers = [ # List of https://pypi.org/classifiers/
 ]
 dependencies = [
   # go/keep-sorted start
-  "PyYAML>=6.0.2, <7.0.0",                          # For APIHubToolset.
-  "absolufy-imports>=0.3.1, <1.0.0",                # For Agent Engine deployment.
-  "anyio>=4.9.0, <5.0.0;python_version>='3.10'",    # For MCP Session Manager
-  "authlib>=1.5.1, <2.0.0",                         # For RestAPI Tool
-  "click>=8.1.8, <9.0.0",                           # For CLI tools
-  "fastapi>=0.115.0, <1.0.0",                       # FastAPI framework
-  "google-api-python-client>=2.157.0, <3.0.0",      # Google API client discovery
-  "google-cloud-bigtable>=2.32.0",                  # For Bigtable database
+  "PyYAML>=6.0.2, <7.0.0",                                  # For APIHubToolset.
+  "absolufy-imports>=0.3.1, <1.0.0",                        # For Agent Engine deployment.
+  "anyio>=4.9.0, <5.0.0;python_version>='3.10'",            # For MCP Session Manager
+  "authlib>=1.5.1, <2.0.0",                                 # For RestAPI Tool
+  "click>=8.1.8, <9.0.0",                                   # For CLI tools
+  "fastapi>=0.115.0, <1.0.0",                               # FastAPI framework
+  "google-api-python-client>=2.157.0, <3.0.0",              # Google API client discovery
+  "google-cloud-bigtable>=2.32.0",                          # For Bigtable database
   "google-cloud-aiplatform[agent_engines]>=1.95.1, <2.0.0", # For VertexAI integrations, e.g. example store.
-  "google-cloud-secret-manager>=2.22.0, <3.0.0",    # Fetching secrets in RestAPI Tool
-  "google-cloud-spanner>=3.56.0, <4.0.0",           # For Spanner database
-  "google-cloud-speech>=2.30.0, <3.0.0",            # For Audio Transcription
-  "google-cloud-storage>=2.18.0, <3.0.0",           # For GCS Artifact service
-  "google-genai>=1.21.1, <2.0.0",                   # Google GenAI SDK
-  "graphviz>=0.20.2, <1.0.0",                       # Graphviz for graph rendering
-  "mcp>=1.8.0, <2.0.0;python_version>='3.10'",      # For MCP Toolset
-  "opentelemetry-api>=1.31.0, <2.0.0",              # OpenTelemetry
+  "google-cloud-secret-manager>=2.22.0, <3.0.0",            # Fetching secrets in RestAPI Tool
+  "google-cloud-spanner>=3.56.0, <4.0.0",                   # For Spanner database
+  "google-cloud-speech>=2.30.0, <3.0.0",                    # For Audio Transcription
+  "google-cloud-storage>=2.18.0, <3.0.0",                   # For GCS Artifact service
+  "google-genai>=1.21.1, <2.0.0",                           # Google GenAI SDK
+  "graphviz>=0.20.2, <1.0.0",                               # Graphviz for graph rendering
+  "mcp>=1.8.0, <2.0.0;python_version>='3.10'",              # For MCP Toolset
+  "opentelemetry-api>=1.31.0, <2.0.0",                      # OpenTelemetry
   "opentelemetry-exporter-gcp-trace>=1.9.0, <2.0.0",
   "opentelemetry-sdk>=1.31.0, <2.0.0",
-  "pydantic>=2.0, <3.0.0",                          # For data validation/models
-  "python-dateutil>=2.9.0.post0, <3.0.0",           # For Vertext AI Session Service
-  "python-dotenv>=1.0.0, <2.0.0",                   # To manage environment variables
+  "pydantic>=2.0, <3.0.0",                                  # For data validation/models
+  "python-dateutil>=2.9.0.post0, <3.0.0",                   # For Vertext AI Session Service
+  "python-dotenv>=1.0.0, <2.0.0",                           # To manage environment variables
   "requests>=2.32.4, <3.0.0",
-  "sqlalchemy>=2.0, <3.0.0",                        # SQL database ORM
-  "starlette>=0.46.2, <1.0.0",                      # For FastAPI CLI
-  "tenacity>=8.0.0, <9.0.0",                        # For Retry management
+  "sqlalchemy-spanner>=1.14.0",                             # Spanner database session service
+  "sqlalchemy>=2.0, <3.0.0",                                # SQL database ORM
+  "starlette>=0.46.2, <1.0.0",                              # For FastAPI CLI
+  "tenacity>=8.0.0, <9.0.0",                                # For Retry management
   "typing-extensions>=4.5, <5",
-  "tzlocal>=5.3, <6.0",                             # Time zone utilities
-  "uvicorn>=0.34.0, <1.0.0",                        # ASGI server for FastAPI
-  "watchdog>=6.0.0, <7.0.0",                        # For file change detection and hot reload
-  "websockets>=15.0.1, <16.0.0",                    # For BaseLlmFlow
+  "tzlocal>=5.3, <6.0",                                     # Time zone utilities
+  "uvicorn>=0.34.0, <1.0.0",                                # ASGI server for FastAPI
+  "watchdog>=6.0.0, <7.0.0",                                # For file change detection and hot reload
+  "websockets>=15.0.1, <16.0.0",                            # For BaseLlmFlow
   # go/keep-sorted end
 ]
 dynamic = ["version"]
@@ -98,18 +99,20 @@ eval = [
 
 test = [
   # go/keep-sorted start
-  "anthropic>=0.43.0",               # For anthropic model tests
+  "a2a-sdk>=0.3.0,<0.4.0;python_version>='3.10'",
+  "anthropic>=0.43.0",                            # For anthropic model tests
   "langchain-community>=0.3.17",
-  # langgraph 0.5 removed langgraph.graph.graph which we depend on
-  "langgraph>=0.2.60, <= 0.4.10",    # For LangGraphAgent
-  "litellm>=1.75.5, <2.0.0",                 # For LiteLLM tests
-  "llama-index-readers-file>=0.4.0", # For retrieval tests
-  "openai>=1.100.2", # For LiteLLM
+  "langgraph>=0.2.60, <= 0.4.10",                 # For LangGraphAgent
+  "litellm>=1.75.5, <2.0.0",                      # For LiteLLM tests
+  "llama-index-readers-file>=0.4.0",              # For retrieval tests
+  "openai>=1.100.2",                              # For LiteLLM
   "pytest-asyncio>=0.25.0",
   "pytest-mock>=3.14.0",
   "pytest-xdist>=3.6.1",
   "pytest>=8.3.4",
   "python-multipart>=0.0.9",
+  "rouge-score>=0.1.2",
+  "tabulate>=0.9.0",
   # go/keep-sorted end
 ]
 

src/google/adk/a2a/utils/agent_to_a2a.py

@@ -39,11 +39,17 @@ from ...memory.in_memory_memory_service import InMemoryMemoryService
 from ...runners import Runner
 from ...sessions.in_memory_session_service import InMemorySessionService
 from ..executor.a2a_agent_executor import A2aAgentExecutor
+from ..experimental import a2a_experimental
 from .agent_card_builder import AgentCardBuilder
 
 
+@a2a_experimental
 def to_a2a(
-    agent: BaseAgent, *, host: str = "localhost", port: int = 8000
+    agent: BaseAgent,
+    *,
+    host: str = "localhost",
+    port: int = 8000,
+    protocol: str = "http",
 ) -> Starlette:
   """Convert an ADK agent to a A2A Starlette application.
 
@@ -51,13 +57,14 @@ def to_a2a(
       agent: The ADK agent to convert
       host: The host for the A2A RPC URL (default: "localhost")
       port: The port for the A2A RPC URL (default: 8000)
+      protocol: The protocol for the A2A RPC URL (default: "http")
 
   Returns:
       A Starlette application that can be run with uvicorn
 
   Example:
       agent = MyAgent()
-      app = to_a2a(agent, host="localhost", port=8000)
+      app = to_a2a(agent, host="localhost", port=8000, protocol="http")
       # Then run with: uvicorn module:app --host localhost --port 8000
   """
   # Set up ADK logging to ensure logs are visible when using uvicorn directly
@@ -87,7 +94,7 @@ def to_a2a(
   )
 
   # Build agent card
-  rpc_url = f"http://{host}:{port}/"
+  rpc_url = f"{protocol}://{host}:{port}/"
   card_builder = AgentCardBuilder(
       agent=agent,
       rpc_url=rpc_url,

src/google/adk/agents/base_agent.py

@@ -181,6 +181,18 @@ class BaseAgent(BaseModel):
 
     cloned_agent = self.model_copy(update=update)
 
+    # If any field is stored as list and not provided in the update, need to
+    # shallow copy it for the cloned agent to avoid sharing the same list object
+    # with the original agent.
+    for field_name in cloned_agent.__class__.model_fields:
+      if field_name == 'sub_agents':
+        continue
+      if update is not None and field_name in update:
+        continue
+      field = getattr(cloned_agent, field_name)
+      if isinstance(field, list):
+        setattr(cloned_agent, field_name, field.copy())
+
     if update is None or 'sub_agents' not in update:
       # If `sub_agents` is not provided in the update, need to recursively clone
       # the sub-agents to avoid sharing the sub-agents with the original agent.

src/google/adk/agents/base_agent_config.py

@@ -79,12 +79,3 @@ Example:
       default=None,
       description='Optional. The after_agent_callbacks of the agent.',
   )
-
-  def to_agent_config(
-      self, custom_agent_config_cls: Type[TBaseAgentConfig]
-  ) -> TBaseAgentConfig:
-    """Converts this config to the concrete agent config type.
-
-    NOTE: this is for ADK framework use only.
-    """
-    return custom_agent_config_cls.model_validate(self.model_dump())

src/google/adk/agents/invocation_context.py

@@ -40,6 +40,25 @@ class LlmCallsLimitExceededError(Exception):
   """Error thrown when the number of LLM calls exceed the limit."""
 
 
+class RealtimeCacheEntry(BaseModel):
+  """Store audio data chunks for caching before flushing."""
+
+  model_config = ConfigDict(
+      arbitrary_types_allowed=True,
+      extra="forbid",
+  )
+  """The pydantic model config."""
+
+  role: str
+  """The role that created this audio data, typically "user" or "model"."""
+
+  data: types.Blob
+  """The audio data chunk."""
+
+  timestamp: float
+  """Timestamp when the audio chunk was received."""
+
+
 class _InvocationCostManager(BaseModel):
   """A container to keep track of the cost of invocation.
 
@@ -156,6 +175,12 @@ class InvocationContext(BaseModel):
   live_session_resumption_handle: Optional[str] = None
   """The handle for live session resumption."""
 
+  input_realtime_cache: Optional[list[RealtimeCacheEntry]] = None
+  """Caches input audio chunks before flushing to session and artifact services."""
+
+  output_realtime_cache: Optional[list[RealtimeCacheEntry]] = None
+  """Caches output audio chunks before flushing to session and artifact services."""
+
   run_config: Optional[RunConfig] = None
   """Configurations for live agents under this invocation."""
 

src/google/adk/agents/parallel_agent.py

@@ -17,11 +17,9 @@
 from __future__ import annotations
 
 import asyncio
-from typing import Any
+import sys
 from typing import AsyncGenerator
 from typing import ClassVar
-from typing import Dict
-from typing import Type
 
 from typing_extensions import override
 
@@ -49,6 +47,70 @@ def _create_branch_ctx_for_sub_agent(
   return invocation_context
 
 
+# TODO - remove once Python <3.11 is no longer supported.
+async def _merge_agent_run_pre_3_11(
+    agent_runs: list[AsyncGenerator[Event, None]],
+) -> AsyncGenerator[Event, None]:
+  """Merges the agent run event generator.
+  This version works in Python 3.9 and 3.10 and uses custom replacement for
+  asyncio.TaskGroup for tasks cancellation and exception handling.
+
+  This implementation guarantees for each agent, it won't move on until the
+  generated event is processed by upstream runner.
+
+  Args:
+      agent_runs: A list of async generators that yield events from each agent.
+
+  Yields:
+      Event: The next event from the merged generator.
+  """
+  sentinel = object()
+  queue = asyncio.Queue()
+
+  def propagate_exceptions(tasks):
+    # Propagate exceptions and errors from tasks.
+    for task in tasks:
+      if task.done():
+        # Ignore the result (None) of correctly finished tasks and re-raise
+        # exceptions and errors.
+        task.result()
+
+  # Agents are processed in parallel.
+  # Events for each agent are put on queue sequentially.
+  async def process_an_agent(events_for_one_agent):
+    try:
+      async for event in events_for_one_agent:
+        resume_signal = asyncio.Event()
+        await queue.put((event, resume_signal))
+        # Wait for upstream to consume event before generating new events.
+        await resume_signal.wait()
+    finally:
+      # Mark agent as finished.
+      await queue.put((sentinel, None))
+
+  tasks = []
+  try:
+    for events_for_one_agent in agent_runs:
+      tasks.append(asyncio.create_task(process_an_agent(events_for_one_agent)))
+
+    sentinel_count = 0
+    # Run until all agents finished processing.
+    while sentinel_count < len(agent_runs):
+      propagate_exceptions(tasks)
+      event, resume_signal = await queue.get()
+      # Agent finished processing.
+      if event is sentinel:
+        sentinel_count += 1
+      else:
+        yield event
+        # Signal to agent that event has been processed by runner and it can
+        # continue now.
+        resume_signal.set()
+  finally:
+    for task in tasks:
+      task.cancel()
+
+
 async def _merge_agent_run(
     agent_runs: list[AsyncGenerator[Event, None]],
 ) -> AsyncGenerator[Event, None]:
@@ -63,30 +125,37 @@ async def _merge_agent_run(
   Yields:
       Event: The next event from the merged generator.
   """
-  tasks = [
-      asyncio.create_task(events_for_one_agent.__anext__())
-      for events_for_one_agent in agent_runs
-  ]
-  pending_tasks = set(tasks)
+  sentinel = object()
+  queue = asyncio.Queue()
 
-  while pending_tasks:
-    done, pending_tasks = await asyncio.wait(
-        pending_tasks, return_when=asyncio.FIRST_COMPLETED
-    )
-    for task in done:
-      try:
-        yield task.result()
+  # Agents are processed in parallel.
+  # Events for each agent are put on queue sequentially.
+  async def process_an_agent(events_for_one_agent):
+    try:
+      async for event in events_for_one_agent:
+        resume_signal = asyncio.Event()
+        await queue.put((event, resume_signal))
+        # Wait for upstream to consume event before generating new events.
+        await resume_signal.wait()
+    finally:
+      # Mark agent as finished.
+      await queue.put((sentinel, None))
 
-        # Find the generator that produced this event and move it on.
-        for i, original_task in enumerate(tasks):
-          if task == original_task:
-            new_task = asyncio.create_task(agent_runs[i].__anext__())
-            tasks[i] = new_task
-            pending_tasks.add(new_task)
-            break  # stop iterating once found
+  async with asyncio.TaskGroup() as tg:
+    for events_for_one_agent in agent_runs:
+      tg.create_task(process_an_agent(events_for_one_agent))
 
-      except StopAsyncIteration:
-        continue
+    sentinel_count = 0
+    # Run until all agents finished processing.
+    while sentinel_count < len(agent_runs):
+      event, resume_signal = await queue.get()
+      # Agent finished processing.
+      if event is sentinel:
+        sentinel_count += 1
+      else:
+        yield event
+        # Signal to agent that it should generate next event.
+        resume_signal.set()
 
 
 class ParallelAgent(BaseAgent):
@@ -112,10 +181,19 @@ class ParallelAgent(BaseAgent):
         )
         for sub_agent in self.sub_agents
     ]
-
-    async with Aclosing(_merge_agent_run(agent_runs)) as agen:
-      async for event in agen:
-        yield event
+    try:
+      # TODO remove if once Python <3.11 is no longer supported.
+      if sys.version_info >= (3, 11):
+        async with Aclosing(_merge_agent_run(agent_runs)) as agen:
+          async for event in agen:
+            yield event
+      else:
+        async with Aclosing(_merge_agent_run_pre_3_11(agent_runs)) as agen:
+          async for event in agen:
+            yield event
+    finally:
+      for sub_agent_run in agent_runs:
+        await sub_agent_run.aclose()
 
   @override
   async def _run_live_impl(

src/google/adk/cli/adk_web_server.py

@@ -45,6 +45,7 @@ from opentelemetry.sdk.trace import TracerProvider
 from pydantic import Field
 from pydantic import ValidationError
 from starlette.types import Lifespan
+from typing_extensions import deprecated
 from typing_extensions import override
 from watchdog.observers import Observer
 
@@ -66,6 +67,7 @@ from ..evaluation.eval_metrics import EvalMetricResult
 from ..evaluation.eval_metrics import EvalMetricResultPerInvocation
 from ..evaluation.eval_metrics import MetricInfo
 from ..evaluation.eval_result import EvalSetResult
+from ..evaluation.eval_set import EvalSet
 from ..evaluation.eval_set_results_manager import EvalSetResultsManager
 from ..evaluation.eval_sets_manager import EvalSetsManager
 from ..events.event import Event
@@ -171,7 +173,18 @@ class AddSessionToEvalSetRequest(common.BaseModel):
 
 
 class RunEvalRequest(common.BaseModel):
-  eval_ids: list[str]  # if empty, then all evals in the eval set are run.
+  eval_ids: list[str] = Field(
+      deprecated=True,
+      default_factory=list,
+      description="This field is deprecated, use eval_case_ids instead.",
+  )
+  eval_case_ids: list[str] = Field(
+      default_factory=list,
+      description=(
+          "List of eval case ids to evaluate. if empty, then all eval cases in"
+          " the eval set are run."
+      ),
+  )
   eval_metrics: list[EvalMetric]
 
 
@@ -193,10 +206,38 @@ class RunEvalResult(common.BaseModel):
   session_id: str
 
 
+class RunEvalResponse(common.BaseModel):
+  run_eval_results: list[RunEvalResult]
+
+
 class GetEventGraphResult(common.BaseModel):
   dot_src: str
 
 
+class CreateEvalSetRequest(common.BaseModel):
+  eval_set: EvalSet
+
+
+class ListEvalSetsResponse(common.BaseModel):
+  eval_set_ids: list[str]
+
+
+class EvalResult(EvalSetResult):
+  """This class has no field intentionally.
+
+  The goal here is to just give a new name to the class to align with the API
+  endpoint.
+  """
+
+
+class ListEvalResultsResponse(common.BaseModel):
+  eval_result_ids: list[str]
+
+
+class ListMetricsInfoResponse(common.BaseModel):
+  metrics_info: list[MetricInfo]
+
+
 class AdkWebServer:
   """Helper class for setting up and running the ADK web server on FastAPI.
 
@@ -466,36 +507,78 @@ class AdkWebServer:
       )
 
     @app.post(
-        "/apps/{app_name}/eval_sets/{eval_set_id}",
+        "/apps/{app_name}/eval-sets",
         response_model_exclude_none=True,
         tags=[TAG_EVALUATION],
     )
     async def create_eval_set(
-        app_name: str,
-        eval_set_id: str,
-    ):
-      """Creates an eval set, given the id."""
+        app_name: str, create_eval_set_request: CreateEvalSetRequest
+    ) -> EvalSet:
       try:
-        self.eval_sets_manager.create_eval_set(app_name, eval_set_id)
+        return self.eval_sets_manager.create_eval_set(
+            app_name=app_name,
+            eval_set_id=create_eval_set_request.eval_set.eval_set_id,
+        )
       except ValueError as ve:
         raise HTTPException(
             status_code=400,
             detail=str(ve),
         ) from ve
 
+    @deprecated(
+        "Please use create_eval_set instead. This will be removed in future"
+        " releases."
+    )
+    @app.post(
+        "/apps/{app_name}/eval_sets/{eval_set_id}",
+        response_model_exclude_none=True,
+        tags=[TAG_EVALUATION],
+    )
+    async def create_eval_set_legacy(
+        app_name: str,
+        eval_set_id: str,
+    ):
+      """Creates an eval set, given the id."""
+      await create_eval_set(
+          app_name=app_name,
+          create_eval_set_request=CreateEvalSetRequest(
+              eval_set=EvalSet(eval_set_id=eval_set_id, eval_cases=[])
+          ),
+      )
+
+    @app.get(
+        "/apps/{app_name}/eval-sets",
+        response_model_exclude_none=True,
+        tags=[TAG_EVALUATION],
+    )
+    async def list_eval_sets(app_name: str) -> ListEvalSetsResponse:
+      """Lists all eval sets for the given app."""
+      eval_sets = []
+      try:
+        eval_sets = self.eval_sets_manager.list_eval_sets(app_name)
+      except NotFoundError as e:
+        logger.warning(e)
+
+      return ListEvalSetsResponse(eval_set_ids=eval_sets)
+
+    @deprecated(
+        "Please use list_eval_sets instead. This will be removed in future"
+        " releases."
+    )
     @app.get(
         "/apps/{app_name}/eval_sets",
         response_model_exclude_none=True,
         tags=[TAG_EVALUATION],
     )
-    async def list_eval_sets(app_name: str) -> list[str]:
-      """Lists all eval sets for the given app."""
-      try:
-        return self.eval_sets_manager.list_eval_sets(app_name)
-      except NotFoundError as e:
-        logger.warning(e)
-        return []
+    async def list_eval_sets_legacy(app_name: str) -> list[str]:
+      list_eval_sets_response = await list_eval_sets(app_name)
+      return list_eval_sets_response.eval_set_ids
 
+    @app.post(
+        "/apps/{app_name}/eval-sets/{eval_set_id}/add-session",
+        response_model_exclude_none=True,
+        tags=[TAG_EVALUATION],
+    )
     @app.post(
         "/apps/{app_name}/eval_sets/{eval_set_id}/add_session",
         response_model_exclude_none=True,
@@ -555,6 +638,11 @@ class AdkWebServer:
 
       return sorted([x.eval_id for x in eval_set_data.eval_cases])
 
+    @app.get(
+        "/apps/{app_name}/eval-sets/{eval_set_id}/eval-cases/{eval_case_id}",
+        response_model_exclude_none=True,
+        tags=[TAG_EVALUATION],
+    )
     @app.get(
         "/apps/{app_name}/eval_sets/{eval_set_id}/evals/{eval_case_id}",
         response_model_exclude_none=True,
@@ -578,6 +666,11 @@ class AdkWebServer:
           ),
       )
 
+    @app.put(
+        "/apps/{app_name}/eval-sets/{eval_set_id}/eval-cases/{eval_case_id}",
+        response_model_exclude_none=True,
+        tags=[TAG_EVALUATION],
+    )
     @app.put(
         "/apps/{app_name}/eval_sets/{eval_set_id}/evals/{eval_case_id}",
         response_model_exclude_none=True,
@@ -610,6 +703,10 @@ class AdkWebServer:
       except NotFoundError as nfe:
         raise HTTPException(status_code=404, detail=str(nfe)) from nfe
 
+    @app.delete(
+        "/apps/{app_name}/eval-sets/{eval_set_id}/eval-cases/{eval_case_id}",
+        tags=[TAG_EVALUATION],
+    )
     @app.delete(
         "/apps/{app_name}/eval_sets/{eval_set_id}/evals/{eval_case_id}",
         tags=[TAG_EVALUATION],
@@ -624,14 +721,30 @@ class AdkWebServer:
       except NotFoundError as nfe:
         raise HTTPException(status_code=404, detail=str(nfe)) from nfe
 
+    @deprecated(
+        "Please use run_eval instead. This will be removed in future releases."
+    )
     @app.post(
         "/apps/{app_name}/eval_sets/{eval_set_id}/run_eval",
         response_model_exclude_none=True,
         tags=[TAG_EVALUATION],
     )
-    async def run_eval(
+    async def run_eval_legacy(
         app_name: str, eval_set_id: str, req: RunEvalRequest
     ) -> list[RunEvalResult]:
+      run_eval_response = await run_eval(
+          app_name=app_name, eval_set_id=eval_set_id, req=req
+      )
+      return run_eval_response.run_eval_results
+
+    @app.post(
+        "/apps/{app_name}/eval-sets/{eval_set_id}/run",
+        response_model_exclude_none=True,
+        tags=[TAG_EVALUATION],
+    )
+    async def run_eval(
+        app_name: str, eval_set_id: str, req: RunEvalRequest
+    ) -> RunEvalResponse:
       """Runs an eval given the details in the eval request."""
       # Create a mapping from eval set file to all the evals that needed to be
       # run.
@@ -661,7 +774,7 @@ class AdkWebServer:
         inference_request = InferenceRequest(
             app_name=app_name,
             eval_set_id=eval_set.eval_set_id,
-            eval_case_ids=req.eval_ids,
+            eval_case_ids=req.eval_case_ids or req.eval_ids,
             inference_config=InferenceConfig(),
         )
         inference_results = await _collect_inferences(
@@ -694,18 +807,41 @@ class AdkWebServer:
             )
         )
 
-      return run_eval_results
+      return RunEvalResponse(run_eval_results=run_eval_results)
 
     @app.get(
-        "/apps/{app_name}/eval_results/{eval_result_id}",
+        "/apps/{app_name}/eval-results/{eval_result_id}",
         response_model_exclude_none=True,
         tags=[TAG_EVALUATION],
     )
     async def get_eval_result(
         app_name: str,
         eval_result_id: str,
-    ) -> EvalSetResult:
+    ) -> EvalResult:
       """Gets the eval result for the given eval id."""
+      try:
+        eval_set_result = self.eval_set_results_manager.get_eval_set_result(
+            app_name, eval_result_id
+        )
+        return EvalResult(**eval_set_result.model_dump())
+      except ValueError as ve:
+        raise HTTPException(status_code=404, detail=str(ve)) from ve
+      except ValidationError as ve:
+        raise HTTPException(status_code=500, detail=str(ve)) from ve
+
+    @deprecated(
+        "Please use get_eval_result instead. This will be removed in future"
+        " releases."
+    )
+    @app.get(
+        "/apps/{app_name}/eval_results/{eval_result_id}",
+        response_model_exclude_none=True,
+        tags=[TAG_EVALUATION],
+    )
+    async def get_eval_result_legacy(
+        app_name: str,
+        eval_result_id: str,
+    ) -> EvalSetResult:
       try:
         return self.eval_set_results_manager.get_eval_set_result(
             app_name, eval_result_id
@@ -715,28 +851,47 @@ class AdkWebServer:
       except ValidationError as ve:
         raise HTTPException(status_code=500, detail=str(ve)) from ve
 
+    @app.get(
+        "/apps/{app_name}/eval-results",
+        response_model_exclude_none=True,
+        tags=[TAG_EVALUATION],
+    )
+    async def list_eval_results(app_name: str) -> ListEvalResultsResponse:
+      """Lists all eval results for the given app."""
+      eval_result_ids = self.eval_set_results_manager.list_eval_set_results(
+          app_name
+      )
+      return ListEvalResultsResponse(eval_result_ids=eval_result_ids)
+
+    @deprecated(
+        "Please use list_eval_results instead. This will be removed in future"
+        " releases."
+    )
     @app.get(
         "/apps/{app_name}/eval_results",
         response_model_exclude_none=True,
         tags=[TAG_EVALUATION],
     )
-    async def list_eval_results(app_name: str) -> list[str]:
-      """Lists all eval results for the given app."""
-      return self.eval_set_results_manager.list_eval_set_results(app_name)
+    async def list_eval_results_legacy(app_name: str) -> list[str]:
+      list_eval_results_response = await list_eval_results(app_name)
+      return list_eval_results_response.eval_result_ids
 
     @app.get(
-        "/apps/{app_name}/eval_metrics",
+        "/apps/{app_name}/metrics-info",
         response_model_exclude_none=True,
         tags=[TAG_EVALUATION],
     )
-    async def list_eval_metrics(app_name: str) -> list[MetricInfo]:
+    async def list_metrics_info(app_name: str) -> ListMetricsInfoResponse:
       """Lists all eval metrics for the given app."""
       try:
         from ..evaluation.metric_evaluator_registry import DEFAULT_METRIC_EVALUATOR_REGISTRY
 
         # Right now we ignore the app_name as eval metrics are not tied to the
         # app_name, but they could be moving forward.
-        return DEFAULT_METRIC_EVALUATOR_REGISTRY.get_registered_metrics()
+        metrics_info = (
+            DEFAULT_METRIC_EVALUATOR_REGISTRY.get_registered_metrics()
+        )
+        return ListMetricsInfoResponse(metrics_info=metrics_info)
       except ModuleNotFoundError as e:
         logger.exception("%s\n%s", MISSING_EVAL_DEPENDENCIES_MESSAGE, e)
         raise HTTPException(

src/google/adk/cli/cli_deploy.py

@@ -13,6 +13,7 @@
 # limitations under the License.
 from __future__ import annotations
 
+import json
 import os
 import shutil
 import subprocess
@@ -28,9 +29,6 @@ WORKDIR /app
 # Create a non-root user
 RUN adduser --disabled-password --gecos "" myuser
 
-# Change ownership of /app to myuser
-RUN chown -R myuser:myuser /app
-
 # Switch to the non-root user
 USER myuser
 
@@ -49,8 +47,8 @@ RUN pip install google-adk=={adk_version}
 
 # Copy agent - Start
 
-COPY "agents/{app_name}/" "/app/agents/{app_name}/"
-{install_agent_deps}
+# Set permission
+COPY --chown=myuser:myuser "agents/{app_name}/" "/app/agents/{app_name}/"
 
 # Copy agent - End
 
@@ -95,6 +93,52 @@ def _resolve_project(project_in_option: Optional[str]) -> str:
   return project
 
 
+def _validate_gcloud_extra_args(
+    extra_gcloud_args: Optional[tuple[str, ...]], adk_managed_args: set[str]
+) -> None:
+  """Validates that extra gcloud args don't conflict with ADK-managed args.
+
+  This function dynamically checks for conflicts based on the actual args
+  that ADK will set, rather than using a hardcoded list.
+
+  Args:
+    extra_gcloud_args: User-provided extra arguments for gcloud.
+    adk_managed_args: Set of argument names that ADK will set automatically.
+                     Should include '--' prefix (e.g., '--project').
+
+  Raises:
+    click.ClickException: If any conflicts are found.
+  """
+  if not extra_gcloud_args:
+    return
+
+  # Parse user arguments into a set of argument names for faster lookup
+  user_arg_names = set()
+  for arg in extra_gcloud_args:
+    if arg.startswith('--'):
+      # Handle both '--arg=value' and '--arg value' formats
+      arg_name = arg.split('=')[0]
+      user_arg_names.add(arg_name)
+
+  # Check for conflicts with ADK-managed args
+  conflicts = user_arg_names.intersection(adk_managed_args)
+
+  if conflicts:
+    conflict_list = ', '.join(f"'{arg}'" for arg in sorted(conflicts))
+    if len(conflicts) == 1:
+      raise click.ClickException(
+          f"The argument {conflict_list} conflicts with ADK's automatic"
+          ' configuration. ADK will set this argument automatically, so please'
+          ' remove it from your command.'
+      )
+    else:
+      raise click.ClickException(
+          f"The arguments {conflict_list} conflict with ADK's automatic"
+          ' configuration. ADK will set these arguments automatically, so'
+          ' please remove them from your command.'
+      )
+
+
 def _get_service_option_by_adk_version(
     adk_version: str,
     session_uri: Optional[str],
@@ -141,6 +185,7 @@ def to_cloud_run(
     artifact_service_uri: Optional[str] = None,
     memory_service_uri: Optional[str] = None,
     a2a: bool = False,
+    extra_gcloud_args: Optional[tuple[str, ...]] = None,
 ):
   """Deploys an agent to Google Cloud Run.
 
@@ -234,26 +279,56 @@ def to_cloud_run(
     click.echo('Deploying to Cloud Run...')
     region_options = ['--region', region] if region else []
     project = _resolve_project(project)
-    subprocess.run(
-        [
-            'gcloud',
-            'run',
-            'deploy',
-            service_name,
-            '--source',
-            temp_folder,
-            '--project',
-            project,
-            *region_options,
-            '--port',
-            str(port),
-            '--verbosity',
-            log_level.lower() if log_level else verbosity,
-            '--labels',
-            'created-by=adk',
-        ],
-        check=True,
-    )
+
+    # Build the set of args that ADK will manage
+    adk_managed_args = {'--source', '--project', '--port', '--verbosity'}
+    if region:
+      adk_managed_args.add('--region')
+
+    # Validate that extra gcloud args don't conflict with ADK-managed args
+    _validate_gcloud_extra_args(extra_gcloud_args, adk_managed_args)
+
+    # Build the command with extra gcloud args
+    gcloud_cmd = [
+        'gcloud',
+        'run',
+        'deploy',
+        service_name,
+        '--source',
+        temp_folder,
+        '--project',
+        project,
+        *region_options,
+        '--port',
+        str(port),
+        '--verbosity',
+        log_level.lower() if log_level else verbosity,
+    ]
+
+    # Handle labels specially - merge user labels with ADK label
+    user_labels = []
+    extra_args_without_labels = []
+
+    if extra_gcloud_args:
+      for arg in extra_gcloud_args:
+        if arg.startswith('--labels='):
+          # Extract user-provided labels
+          user_labels_value = arg[9:]  # Remove '--labels=' prefix
+          user_labels.append(user_labels_value)
+        else:
+          extra_args_without_labels.append(arg)
+
+    # Combine ADK label with user labels
+    all_labels = ['created-by=adk']
+    all_labels.extend(user_labels)
+    labels_arg = ','.join(all_labels)
+
+    gcloud_cmd.extend(['--labels', labels_arg])
+
+    # Add any remaining extra passthrough args
+    gcloud_cmd.extend(extra_args_without_labels)
+
+    subprocess.run(gcloud_cmd, check=True)
   finally:
     click.echo(f'Cleaning up the temp folder: {temp_folder}')
     shutil.rmtree(temp_folder)
@@ -274,6 +349,7 @@ def to_agent_engine(
     description: Optional[str] = None,
     requirements_file: Optional[str] = None,
     env_file: Optional[str] = None,
+    agent_engine_config_file: Optional[str] = None,
 ):
   """Deploys an agent to Vertex AI Agent Engine.
 
@@ -323,6 +399,9 @@ def to_agent_engine(
       variables. If not specified, the `.env` file in the `agent_folder` will be
       used. The values of `GOOGLE_CLOUD_PROJECT` and `GOOGLE_CLOUD_LOCATION`
       will be overridden by `project` and `region` if they are specified.
+    agent_engine_config_file (str): The filepath to the agent engine config file
+      to use. If not specified, the `.agent_engine_config.json` file in the
+      `agent_folder` will be used.
   """
   app_name = os.path.basename(agent_folder)
   agent_src_path = os.path.join(temp_folder, app_name)
@@ -353,6 +432,34 @@ def to_agent_engine(
     project = _resolve_project(project)
 
     click.echo('Resolving files and dependencies...')
+    agent_config = {}
+    if not agent_engine_config_file:
+      # Attempt to read the agent engine config from .agent_engine_config.json in the dir (if any).
+      agent_engine_config_file = os.path.join(
+          agent_folder, '.agent_engine_config.json'
+      )
+    if os.path.exists(agent_engine_config_file):
+      click.echo(f'Reading agent engine config from {agent_engine_config_file}')
+      with open(agent_engine_config_file, 'r') as f:
+        agent_config = json.load(f)
+    if display_name:
+      if 'display_name' in agent_config:
+        click.echo(
+            'Overriding display_name in agent engine config with'
+            f' {display_name}'
+        )
+      agent_config['display_name'] = display_name
+    if description:
+      if 'description' in agent_config:
+        click.echo(
+            f'Overriding description in agent engine config with {description}'
+        )
+      agent_config['description'] = description
+    if agent_config.get('extra_packages'):
+      agent_config['extra_packages'].append(temp_folder)
+    else:
+      agent_config['extra_packages'] = [temp_folder]
+
     if not requirements_file:
       # Attempt to read requirements from requirements.txt in the dir (if any).
       requirements_txt_path = os.path.join(agent_src_path, 'requirements.txt')
@@ -361,7 +468,18 @@ def to_agent_engine(
         with open(requirements_txt_path, 'w', encoding='utf-8') as f:
           f.write('google-cloud-aiplatform[adk,agent_engines]')
         click.echo(f'Created {requirements_txt_path}')
-      requirements_file = requirements_txt_path
+      agent_config['requirements'] = agent_config.get(
+          'requirements',
+          requirements_txt_path,
+      )
+    else:
+      if 'requirements' in agent_config:
+        click.echo(
+            'Overriding requirements in agent engine config with '
+            f'{requirements_file}'
+        )
+      agent_config['requirements'] = requirements_file
+
     env_vars = None
     if not env_file:
       # Attempt to read the env variables from .env in the dir (if any).
@@ -395,6 +513,14 @@ def to_agent_engine(
           else:
             region = env_region
             click.echo(f'{region=} set by GOOGLE_CLOUD_LOCATION in {env_file}')
+    if env_vars:
+      if 'env_vars' in agent_config:
+        click.echo(
+            f'Overriding env_vars in agent engine config with {env_vars}'
+        )
+      agent_config['env_vars'] = env_vars
+    # Set env_vars in agent_config to None if it is not set.
+    agent_config['env_vars'] = agent_config.get('env_vars', env_vars)
 
     vertexai.init(
         project=project,
@@ -406,7 +532,7 @@ def to_agent_engine(
     is_config_agent = False
     config_root_agent_file = os.path.join(agent_src_path, 'root_agent.yaml')
     if os.path.exists(config_root_agent_file):
-      click.echo('Config agent detected.')
+      click.echo(f'Config agent detected: {config_root_agent_file}')
       is_config_agent = True
 
     adk_app_file = os.path.join(temp_folder, f'{adk_app}.py')
@@ -439,7 +565,7 @@ def to_agent_engine(
               click.echo(f'The following exception was raised: {e}')
 
     click.echo('Deploying to agent engine...')
-    agent_engine = agent_engines.ModuleAgent(
+    agent_config['agent_engine'] = agent_engines.ModuleAgent(
         module_name=adk_app,
         agent_name='adk_app',
         register_operations={
@@ -461,14 +587,6 @@ def to_agent_engine(
         sys_paths=[temp_folder[1:]],
         agent_framework='google-adk',
     )
-    agent_config = dict(
-        agent_engine=agent_engine,
-        requirements=requirements_file,
-        display_name=display_name,
-        description=description,
-        env_vars=env_vars,
-        extra_packages=[temp_folder],
-    )
 
     if not agent_engine_id:
       agent_engines.create(**agent_config)