Merge https://github.com/google/adk-python/pull/4025
**Please ensure you have read the [contribution guide](https://github.com/google/adk-python/blob/main/CONTRIBUTING.md) before creating a pull request.**
### Link to Issue or Description of Change
**1. Link to an existing issue (if applicable):**
- Closes:
- #3950
- #3731
- #3708
**2. Or, if no issue exists, describe the change:**
**Problem:**
- `ClientSession` of https://github.com/modelcontextprotocol/python-sdk uses AnyIO for async task management.
- AnyIO TaskGroup requires its start and close must happen in a same task.
- Since `McpSessionManager` does not create task per client, the client might be closed by different task, cause the error: `Attempted to exit cancel scope in a different task than it was entered in`.
**Solution:**
I Suggest 2 changes:
Handling the `ClientSession` in a single task
- To start and close `ClientSession` by the same task, we need to wrap the whole lifecycle of `ClientSession` to a single task.
- `SessionContext` wraps the initialization and disposal of `ClientSession` to a single task, ensures that the `ClientSession` will be handled only in a dedicated task.
Add timeout for `ClientSession`
- Since now we are using task per `ClientSession`, task should never be leaked.
- But `McpSessionManager` does not deliver timeout directly to `ClientSession` when the type is not STDIO.
- There is only timeout for `httpx` client when MCP type is SSE or StreamableHTTP.
- But the timeout applys only to `httpx` client, so if there is an issue in MCP client itself(e.g. https://github.com/modelcontextprotocol/python-sdk/issues/262), a tool call waits the result **FOREVER**!
- To overcome this issue, I propagated the `sse_read_timeout` to `ClientSession`.
- `timeout` is too short for timeout for tool call, since its default value is only 5s.
- `sse_read_timeout` is originally made for read timeout of SSE(default value of 5m or 300s), but actually most of SSE implementations from server (e.g. FastAPI, etc.) sends ping periodically(about 15s I assume), so in a normal circumstances this timeout is quite useless.
- If the server does not send ping, the timeout is equal to tool call timeout. Therefore, it would be appropriate to use `sse_read_timeout` as tool call timeout.
- Most of tool calls should finish within 5 minutes, and sse timeout is adjustable if not.
- If this change is not acceptable, we could make a dedicate parameter for tool call timeout(e.g. `tool_call_timeout`).
### Testing Plan
- Although this does not change the interface itself, it changes its own session management logics, some existing tests are no longer valid.
- I made changes to those tests, especially those of which validate session states(e.g. checking whether `initialize()` called).
- Since now session is encapsulated with `SessionContext`, we cannot validate the initialized state of the session in `TestMcpSessionManager`, should validate it at `TestSessionContext`.
- Added a simple test for reproducing the issue(`test_create_and_close_session_in_different_tasks`).
- Also made a test for the new component: `SessionContext`.
**Unit Tests:**
- [x] I have added or updated unit tests for my change.
- [x] All unit tests pass locally.
```plaintext
=================================================================================== 3689 passed, 1 skipped, 2205 warnings in 63.39s (0:01:03) ===================================================================================
```
**Manual End-to-End (E2E) Tests:**
_Please provide instructions on how to manually test your changes, including any
necessary setup or configuration. Please provide logs or screenshots to help
reviewers better understand the fix._
### Checklist
- [x] I have read the [CONTRIBUTING.md](https://github.com/google/adk-python/blob/main/CONTRIBUTING.md) document.
- [x] I have performed a self-review of my own code.
- [x] I have commented my code, particularly in hard-to-understand areas.
- [x] I have added tests that prove my fix is effective or that my feature works.
- [x] New and existing unit tests pass locally with my changes.
- [x] I have manually tested my changes end-to-end.
- [ ] ~~Any dependent changes have been merged and published in downstream modules.~~ `no deps has been changed`
### Additional context
This PR is related to https://github.com/modelcontextprotocol/python-sdk/pull/1817 since it also fixes endless tool call awaiting.
Co-authored-by: Kathy Wu <wukathy@google.com>
COPYBARA_INTEGRATE_REVIEW=https://github.com/google/adk-python/pull/4025 from challenger71498:feat/task-based-mcp-session-manager f7f7cd0c9c96840361c30499d08c33a189f57d86
PiperOrigin-RevId: 856438147
1. Convert A2A responses containing a DataPart to ADK events. By default, this is done by serializing the DataPart to JSON and embedding it within the inline_data field of a GenAI Part, wrapped with custom tags (<a2a_datapart_json> and </a2a_datapart_json>).
2. Convert ADK events back to A2A requests. Specifically, messages stored in inline_data with the text/plain mime type and content wrapped within the custom tags (<a2a_datapart_json> and </a2a_datapart_json>) are deserialized from JSON back into an A2A DataPart
PiperOrigin-RevId: 856426615
audio is transcribed thus no need to be sent, but other blob(e.g. image) should still be sent.
Co-authored-by: Xiang (Sean) Zhou <seanzhougoogle@google.com>
PiperOrigin-RevId: 856422986
Introduces a function to map LiteLLM finish reason strings to the internal types.FinishReason enum and populates the finish_reason field in LlmResponse. Removes custom logic for handling file URIs, including special casing for different providers, and updates tests accordingly
Close#4125
Co-authored-by: George Weale <gweale@google.com>
PiperOrigin-RevId: 856421317
There was an extra test_mcp_toolset in the tools/ directory with only one test; I moved it into the main file.
Co-authored-by: Kathy Wu <wukathy@google.com>
PiperOrigin-RevId: 856419611
The LoadArtifactsTool now checks if an artifact's inline data MIME type is supported by Gemini. If not, it attempts to convert the artifact content into a text Part
Close#4028
Co-authored-by: George Weale <gweale@google.com>
PiperOrigin-RevId: 856404510
Explicitly pass the log_level parameter to uvicorn.Config in both
adk web and adk api_server commands. This ensures that Uvicorn's
internal logging respects the configured log level.
Closes#4139
feature/auto-create-new-session
Merge https://github.com/google/adk-python/pull/4072
**Please ensure you have read the [contribution guide](https://github.com/google/adk-python/blob/main/CONTRIBUTING.md) before creating a pull request.**
### Link to Issue or Description of Change
**2. Or, if no issue exists, describe the change:**
**Problem:**
When building frontend applications with ADK, there's a limitation where frontends cannot always guarantee that `create_session` is called before initiating a conversation. This creates friction in the user experience because:
- Users may refresh the page or navigate directly to a conversation URL with a specific session_id
- Frontend state management may lose track of whether a session was already created
- Mobile apps or single-page applications have complex lifecycle management where ensuring `create_session` is called first adds unnecessary complexity
- This forces developers to implement additional logic to check session existence before every conversation
Currently, if `get_session` is called with a non-existent session_id, it returns `None`, requiring the frontend to explicitly handle this case and call `create_session` separately.
**Solution:**
Modified the `get_session` method in `DatabaseSessionService` to automatically create a session if it doesn't exist in the database. This "get or create" pattern is common in many frameworks and provides a more developer-friendly API.
The implementation:
1. Attempts to fetch the session from the database
2. If the session doesn't exist (returns `None`), automatically calls `create_session` with the provided parameters
3. Retrieves and returns the newly created session
4. Maintains backward compatibility - existing code continues to work without changes
This allows frontends to simply call `get_session` with a session_id and be confident that the session will be available, regardless of whether it was previously created.
**Benefits:**
- Simplifies frontend integration by removing the need to track session creation state
- Reduces API calls (no need to check existence before calling get_session)
- Follows the principle of least surprise - getting a session with an ID should work reliably
- No breaking changes to existing code that checks for `None` return values
### Testing Plan
**Unit Tests:**
- [x] I have added or updated unit tests for my change.
- [x] All unit tests pass locally.
**pytest results:**
COPYBARA_INTEGRATE_REVIEW=https://github.com/google/adk-python/pull/4072 from lwangverizon:feature/auto-create-new-session 5475c6ae91d12332598d521302736eb1db79a8be
PiperOrigin-RevId: 856019482
LlmResponse so far doesn't expose generation_complete signal, removing the dead codes.
Co-authored-by: Xiang (Sean) Zhou <seanzhougoogle@google.com>
PiperOrigin-RevId: 855835802
### Description of Change
**Problem:**
The `ToolboxToolset` was implemented directly within `adk-python`, leading to code duplication and potential drift from the core `toolbox-adk` implementation.
**Solution:**
Refactored `ToolboxToolset` to act as a rigorous wrapper around the `toolbox-adk` package, which delegates all functionality to `toolbox_adk.ToolboxToolset`.
### Testing Plan
**Unit Tests:**
- [x] I have added or updated unit tests for my change.
- [x] All unit tests pass locally.
Summary:
- Verified initialization flows through to `toolbox-adk`.
- Verified `auth_token_getters` are correctly propagated.
- Verified type hints are static-analysis friendly.
**Manual End-to-End (E2E) Tests:**
Manually verified standard toolbox loading and execution with the new wrapper:
```python
from google.adk.tools import ToolboxToolset
from toolbox_adk import CredentialStrategy
# Loading with toolset_name
ts = ToolboxToolset(
server_url='http://localhost:8080',
toolset_name='calculator',
credentials=CredentialStrategy.toolbox_identity()
)
tools = await ts.get_tools()
print(f'Loaded {len(tools)} tools')
```
### Checklist
- [x] I have read the [CONTRIBUTING.md](https://github.com/google/adk-python/blob/main/CONTRIBUTING.md) document.
- [x] I have performed a self-review of my own code.
- [x] I have commented my code, particularly in hard-to-understand areas.
- [x] I have added tests that prove my fix is effective or that my feature works.
- [x] New and existing unit tests pass locally with my changes.
- [x] I have manually tested my changes end-to-end.
- [x] Any dependent changes have been merged and published in downstream modules.
PiperOrigin-RevId: 855798474
Merge https://github.com/google/adk-python/pull/3756
move event iteration inside api_client context in get_session
Move event iteration inside the api_client context manager in VertexAiSessionService.get_session() to prevent client closure during multi-page event fetching.
**Please ensure you have read the [contribution guide](https://github.com/google/adk-python/blob/main/CONTRIBUTING.md) before creating a pull request.**
### Link to Issue or Description of Change
**1. Link to an existing issue (if applicable):**
- Closes: #3757
**2. Or, if no issue exists, describe the change:**
**Problem:**
When a session contains more than 100 events (requiring pagination), `VertexAiSessionService.get_session()` fails with:
```
RuntimeError: Cannot send a request, as the client has been closed.
```
The root cause is that the `events_iterator` is consumed **outside** the `async with self._get_api_client() as api_client:` context block. When the iterator needs to fetch page 2, 3, etc., the API client has already been closed because the `async with` block has exited.
```python
# Current buggy flow:
async with self._get_api_client() as api_client:
get_session_response, events_iterator = await asyncio.gather(...)
# ← Client closed here
async for event in events_iterator: # ← Fails on page 2+ (client closed)
session.events.append(...)
```
**Solution:**
Move the session creation, user validation, and event iteration **inside** the `async with` block so the API client remains open during the entire pagination process:
```python
async with self._get_api_client() as api_client:
get_session_response, events_iterator = await asyncio.gather(...)
# Validation and session creation...
async for event in events_iterator: # ← Now works for all pages
session.events.append(...)
# Client closed after all events are fetched
```
### Testing Plan
**Unit Tests:**
- [x] I have added or updated unit tests for my change.
- [x] All unit tests pass locally.
```bash
pytest tests/unittests/sessions/test_vertex_ai_session_service.py -v
```
**Added regression test:** `test_get_session_pagination_keeps_client_open`
- Creates a `MockAsyncClientWithPagination` that tracks whether it's inside the `async with` context
- Raises `RuntimeError` if iteration happens outside the context (matching real httpx behavior)
- Simulates 3 pages of events (100 + 100 + 50 = 250 events)
- Verifies all 250 events are successfully retrieved
**Manual End-to-End (E2E) Tests:**
1. Deploy an ADK agent to Vertex AI Agent Engine
2. Create a session and send 100+ messages to accumulate >100 events
3. Verify `get_session()` successfully retrieves all events without error
**Before fix:**
```
RuntimeError: Cannot send a request, as the client has been closed.
```
**After fix:**
- Session with 201 events (3 pages) loads successfully
- All events are retrieved and appended to the session
### Checklist
- [x] I have read the [CONTRIBUTING.md](https://github.com/google/adk-python/blob/main/CONTRIBUTING.md) document.
- [x] I have performed a self-review of my own code.
- [x] I have commented my code, particularly in hard-to-understand areas.
- [x] I have added tests that prove my fix is effective or that my feature works.
- [x] New and existing unit tests pass locally with my changes.
- [x] I have manually tested my changes end-to-end.
- [x] Any dependent changes have been merged and published in downstream modules.
### Additional context
This bug affects any production deployment where users have extended conversations. Sessions accumulating >100 events (which triggers pagination) become completely unusable as the agent cannot load the session to process new messages.
The fix is minimal and maintains backward compatibility - it only changes the scope of the `async with` block without altering any logic or return values.
**Affected versions:** Tested on google-adk 1.19.0, but the bug exists in earlier versions as well.
COPYBARA_INTEGRATE_REVIEW=https://github.com/google/adk-python/pull/3756 from AlexisMarasigan:fix/vertex-ai-session-service-paginatio 01fbafa6524312f24f7c9feaffb07bff0ad49b77
PiperOrigin-RevId: 855451813
* **Async Safety:** Improved TraceManager context variable handling to ensure correct context isolation in concurrent asynchronous operations. This was achieved by using immutable tuples for the span stack and making copies of context dictionaries before modification.
* **Enhanced Logging:** The BigQueryAgentAnalyticsPlugin now captures richer metadata, including:
* Root agent name (via a new context variable).
* LLM model name and version.
* Usage metadata from LLM requests and responses.
* **Serialization Fix:** Updated BigQueryAgentAnalyticsPlugin to prevent JSON serialization errors when logging custom objects (e.g., Dataclasses). These are now automatically converted to dictionaries or string representations to ensure successful insertion into BigQuery.
PiperOrigin-RevId: 855415320
Kathy Wu
Google Team Member
Xiang (Sean) Zhou
George Weale
Xuan Yang
Kotaro Saito
lwangverizon
Joseph Pagadora
Liang Wu
Alexis Marasigan