fix(tools): Add proper cleanup for AgentTool to prevent MCP session errors

Clean up runner resources in AgentTool after sub-agent execution to ensure
MCP sessions are closed in the correct async task context. Without this fix,
MCP sessions were cleaned up during garbage collection in a different task,
causing "Attempted to exit cancel scope in a different task" errors.

This fix ensures that `runner.close()` is called immediately after the
sub-agent finishes executing, properly closing all MCP sessions and other
resources within the same async task context they were created in.

Also adds two demo agents showing how to use AgentTool with MCP tools:
- mcp_in_agent_tool_remote: Uses SSE mode (remote server connection)
- mcp_in_agent_tool_stdio: Uses stdio mode (local subprocess)

Both demos use Gemini 2.5 Flash and include zero-installation setup using uvx.

Related: #1112, #929

contributing/samples/live_bidi_streaming_single_agent/agent.py

@@ -65,8 +65,9 @@ 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-live-2.5-flash-preview-native-audio-09-2025',  # vertex
+    # model='gemini-2.5-flash-native-audio-preview-09-2025',  # for AI studio
+    # key
     name='roll_dice_agent',
     description=(
         'hello world agent that can roll a dice of 6 sides and check prime'

contributing/samples/mcp_in_agent_tool_remote/README.md

@@ -0,0 +1,70 @@
+# AgentTool with MCP Demo (SSE Mode)
+
+This demo shows how `AgentTool` works with MCP (Model Context Protocol) toolsets using **SSE mode**.
+
+## SSE vs Stdio Mode
+
+This demo uses **SSE (Server-Sent Events) mode** where the MCP server runs as a separate HTTP server:
+- **Remote connection** - Connects to server via HTTP
+- **Separate process** - Server must be started manually
+- **Network communication** - Uses HTTP/SSE for messaging
+
+For the **stdio (subprocess) version**, see [mcp_in_agent_tool_stdio](../mcp_in_agent_tool_stdio/).
+
+## Setup
+
+**Start the MCP simple-tool server in SSE mode** (in a separate terminal):
+```bash
+# Run the server using uvx (no installation needed)
+# Port 3000 avoids conflict with adk web (which uses 8000)
+uvx --from 'git+https://github.com/modelcontextprotocol/python-sdk.git#subdirectory=examples/servers/simple-tool' \
+    mcp-simple-tool --transport sse --port 3000
+```
+
+The server should be accessible at `http://localhost:3000/sse`.
+
+## Running the Demo
+
+```bash
+adk web contributing/samples
+```
+
+Then select **mcp_in_agent_tool_remote** from the list and interact with the agent.
+
+## Try These Prompts
+
+This demo uses **Gemini 2.5 Flash** as the model. Try these prompts:
+
+1. **Check available tools:**
+   ```
+   What tools do you have access to?
+   ```
+
+2. **Fetch and summarize JSON Schema specification:**
+   ```
+   Use the mcp_helper to fetch https://json-schema.org/specification and summarize the key features of JSON Schema
+   ```
+
+## Architecture
+
+```
+main_agent (root_agent)
+  │
+  └── AgentTool wrapping:
+        │
+        └── mcp_helper (sub_agent)
+              │
+              └── McpToolset (SSE connection)
+                    │
+                    └── http://localhost:3000/sse
+                          │
+                          └── MCP simple-tool server
+                                │
+                                └── Website Fetcher Tool
+```
+
+## Related
+
+- **Issue:** [#1112 - Using agent as tool outside of adk web doesn't exit cleanly](https://github.com/google/adk-python/issues/1112)
+- **Related Issue:** [#929 - LiteLLM giving error with OpenAI models and Grafana's MCP server](https://github.com/google/adk-python/issues/929)
+- **Stdio Version:** [mcp_in_agent_tool_stdio](../mcp_in_agent_tool_stdio/) - Uses local subprocess connection

contributing/samples/mcp_in_agent_tool_remote/__init__.py

@@ -0,0 +1,15 @@
+# Copyright 2025 Google LLC
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from . import agent

contributing/samples/mcp_in_agent_tool_remote/agent.py

@@ -0,0 +1,70 @@
+# Copyright 2025 Google LLC
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from google.adk.agents import Agent
+from google.adk.tools import AgentTool
+from google.adk.tools.mcp_tool import McpToolset
+from google.adk.tools.mcp_tool.mcp_session_manager import SseConnectionParams
+
+# Create MCP toolset
+# This uses the simple-tool MCP server via SSE
+# You need to start the MCP server separately (see README.md)
+mcp_toolset = McpToolset(
+    connection_params=SseConnectionParams(
+        url="http://localhost:3000/sse",
+        timeout=10.0,
+        sse_read_timeout=300.0,
+    )
+)
+
+# Create sub-agent with MCP tools
+# This agent has direct access to MCP tools
+sub_agent = Agent(
+    name="mcp_helper",
+    model="gemini-2.5-flash",
+    description=(
+        "A helpful assistant with access to MCP tools for fetching websites."
+    ),
+    instruction="""You are a helpful assistant with access to MCP tools.
+
+When the user asks for help:
+1. Explain what tools you have available (website fetching)
+2. Use the appropriate tool if needed
+3. Provide clear and helpful responses
+
+You have access to a website fetcher tool via MCP. Use it to fetch and return website content.""",
+    tools=[mcp_toolset],
+)
+
+# Wrap sub-agent as an AgentTool
+# This allows the main agent to delegate tasks to the sub-agent
+# The sub-agent has access to MCP tools for fetching websites
+mcp_agent_tool = AgentTool(agent=sub_agent)
+
+# Create main agent
+# This agent can delegate to the sub-agent via AgentTool
+root_agent = Agent(
+    name="main_agent",
+    model="gemini-2.5-flash",
+    description="Main agent that can delegate to a sub-agent with MCP tools.",
+    instruction="""You are a helpful assistant. You have access to a sub-agent (mcp_helper)
+that has MCP tools for fetching websites.
+
+When the user asks for help:
+- If they need to fetch a website, call the mcp_helper tool
+- Otherwise, respond directly
+
+Always be helpful and explain what you're doing.""",
+    tools=[mcp_agent_tool],
+)

contributing/samples/mcp_in_agent_tool_stdio/README.md

@@ -0,0 +1,70 @@
+# AgentTool with MCP Demo (Stdio Mode)
+
+This demo shows how `AgentTool` works with MCP (Model Context Protocol) toolsets using **stdio mode**.
+
+## Stdio vs SSE Mode
+
+This demo uses **stdio mode** where the MCP server runs as a subprocess:
+- **Simpler setup** - No need to start a separate server
+- **Auto-launched** - Server starts automatically when agent runs
+- **Local process** - Uses stdin/stdout for communication
+
+For the **SSE (remote server) version**, see [mcp_in_agent_tool_remote](../mcp_in_agent_tool_remote/).
+
+## Setup
+
+**No installation required!** The MCP server will be launched automatically using `uvx` when you run the agent.
+
+The demo uses `uvx` to fetch and run the MCP simple-tool server directly from the GitHub repository's subdirectory:
+```bash
+uvx --from 'git+https://github.com/modelcontextprotocol/python-sdk.git#subdirectory=examples/servers/simple-tool' \
+    mcp-simple-tool
+```
+
+This happens automatically via the stdio connection when the agent starts.
+
+## Running the Demo
+
+```bash
+adk web contributing/samples
+```
+
+Then select **mcp_in_agent_tool_stdio** from the list and interact with the agent.
+
+## Try These Prompts
+
+This demo uses **Gemini 2.5 Flash** as the model. Try these prompts:
+
+1. **Check available tools:**
+   ```
+   What tools do you have access to?
+   ```
+
+2. **Fetch and summarize JSON Schema specification:**
+   ```
+   Use the mcp_helper to fetch https://json-schema.org/specification and summarize the key features of JSON Schema
+   ```
+
+## Architecture
+
+```
+main_agent (root_agent)
+  │
+  └── AgentTool wrapping:
+        │
+        └── mcp_helper (sub_agent)
+              │
+              └── McpToolset (stdio connection)
+                    │
+                    └── MCP Server (subprocess via uvx)
+                          │
+                          └── uvx --from git+...#subdirectory=... mcp-simple-tool
+                                │
+                                └── Website Fetcher Tool
+```
+
+## Related
+
+- **Issue:** [#1112 - Using agent as tool outside of adk web doesn't exit cleanly](https://github.com/google/adk-python/issues/1112)
+- **Related Issue:** [#929 - LiteLLM giving error with OpenAI models and Grafana's MCP server](https://github.com/google/adk-python/issues/929)
+- **SSE Version:** [mcp_in_agent_tool_remote](../mcp_in_agent_tool_remote/) - Uses remote server connection

contributing/samples/mcp_in_agent_tool_stdio/__init__.py

@@ -0,0 +1,15 @@
+# Copyright 2025 Google LLC
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from . import agent

contributing/samples/mcp_in_agent_tool_stdio/agent.py

@@ -0,0 +1,77 @@
+# Copyright 2025 Google LLC
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from google.adk.agents import Agent
+from google.adk.tools import AgentTool
+from google.adk.tools.mcp_tool import McpToolset
+from google.adk.tools.mcp_tool.mcp_session_manager import StdioConnectionParams
+from mcp import StdioServerParameters
+
+# Create MCP toolset
+# This uses the simple-tool MCP server via stdio
+# The server will be launched automatically using uvx from the subdirectory
+mcp_toolset = McpToolset(
+    connection_params=StdioConnectionParams(
+        server_params=StdioServerParameters(
+            command="uvx",
+            args=[
+                "--from",
+                "git+https://github.com/modelcontextprotocol/python-sdk.git#subdirectory=examples/servers/simple-tool",
+                "mcp-simple-tool",
+            ],
+        ),
+        timeout=10.0,
+    )
+)
+
+# Create sub-agent with MCP tools
+# This agent has direct access to MCP tools
+sub_agent = Agent(
+    name="mcp_helper",
+    model="gemini-2.5-flash",
+    description=(
+        "A helpful assistant with access to MCP tools for fetching websites."
+    ),
+    instruction="""You are a helpful assistant with access to MCP tools.
+
+When the user asks for help:
+1. Explain what tools you have available (website fetching)
+2. Use the appropriate tool if needed
+3. Provide clear and helpful responses
+
+You have access to a website fetcher tool via MCP. Use it to fetch and return website content.""",
+    tools=[mcp_toolset],
+)
+
+# Wrap sub-agent as an AgentTool
+# This allows the main agent to delegate tasks to the sub-agent
+# The sub-agent has access to MCP tools for fetching websites
+mcp_agent_tool = AgentTool(agent=sub_agent)
+
+# Create main agent
+# This agent can delegate to the sub-agent via AgentTool
+root_agent = Agent(
+    name="main_agent",
+    model="gemini-2.5-flash",
+    description="Main agent that can delegate to a sub-agent with MCP tools.",
+    instruction="""You are a helpful assistant. You have access to a sub-agent (mcp_helper)
+that has MCP tools for fetching websites.
+
+When the user asks for help:
+- If they need to fetch a website, call the mcp_helper tool
+- Otherwise, respond directly
+
+Always be helpful and explain what you're doing.""",
+    tools=[mcp_agent_tool],
+)

src/google/adk/models/gemini_llm_connection.py

@@ -105,13 +105,14 @@ class GeminiLlmConnection(BaseLlmConnection):
     """
     if isinstance(input, types.Blob):
       input_blob = input.model_dump()
-      logger.debug('Sending LLM Blob: %s', input_blob)
+      # The blob is binary and is very large. So let's not log it.
+      logger.debug('Sending LLM Blob.')
       await self._gemini_session.send(input=input_blob)
     elif isinstance(input, types.ActivityStart):
-      logger.debug('Sending LLM activity start signal')
+      logger.debug('Sending LLM activity start signal.')
       await self._gemini_session.send_realtime_input(activity_start=input)
     elif isinstance(input, types.ActivityEnd):
-      logger.debug('Sending LLM activity end signal')
+      logger.debug('Sending LLM activity end signal.')
       await self._gemini_session.send_realtime_input(activity_end=input)
     else:
       raise ValueError('Unsupported input type: %s' % type(input))

src/google/adk/tools/agent_tool.py

@@ -164,6 +164,10 @@ class AgentTool(BaseTool):
         if event.content:
           last_content = event.content
 
+    # Clean up runner resources (especially MCP sessions)
+    # to avoid "Attempted to exit cancel scope in a different task" errors
+    await runner.close()
+
     if not last_content:
       return ''
     merged_text = '\n'.join(p.text for p in last_content.parts if p.text)

tests/unittests/tools/test_agent_tool.py

@@ -131,6 +131,10 @@ async def test_agent_tool_inherits_parent_app_name(monkeypatch):
       )
       return _empty_async_generator()
 
+    async def close(self):
+      """Mock close method."""
+      pass
+
   monkeypatch.setattr('google.adk.runners.Runner', StubRunner)
 
   tool_agent = Agent(