mirror of https://github.com/encounter/adk-python.git synced 2026-07-09 18:19:28 -07:00

Files

T

Wei Sun (Jack) fa5c546a55 fix(tools): Add proper cleanup for AgentTool to prevent MCP session errors

Merge https://github.com/google/adk-python/pull/3411

## Summary

Fixes AgentTool cleanup to prevent MCP session errors by calling `runner.close()` after sub-agent execution.

## Problem

When using AgentTool with MCP tools, the runner cleanup happened during garbage collection in a different async task context, causing:
```
RuntimeError: Attempted to exit cancel scope in a different task than it was entered in
```

## Solution

- Call `await runner.close()` immediately after sub-agent execution in AgentTool
- This ensures MCP sessions and other resources are cleaned up in the correct async task context
- Updated test mock to include the close() method

## Demo Agents

Added two comprehensive demo agents showing how to use AgentTool with MCP tools:

### mcp_in_agent_tool_remote (SSE mode)
- Uses HTTP/SSE connection to remote MCP server
- Zero-installation setup with `uvx`
- Demonstrates server-side MCP deployment pattern

### mcp_in_agent_tool_stdio (stdio mode)
- Uses subprocess connection with automatic server launch
- Fully automatic setup with `uvx` subdirectory syntax
- Demonstrates embedded MCP deployment pattern

Both demos:
- Use Gemini 2.5 Flash
- Include example prompts for JSON Schema exploration
- Have comprehensive READMEs with architecture diagrams
- Follow ADK agent structure conventions

## Testing

- ✅ All existing unit tests pass
- ✅ Manual testing with both SSE and stdio modes
- ✅ Verified cleanup happens in correct async context
- ✅ No more cancel scope errors with MCP tools

## Related

- Fixes #1112
- Related to #929

Co-authored-by: Wei Sun (Jack) <weisun@google.com>
COPYBARA_INTEGRATE_REVIEW=https://github.com/google/adk-python/pull/3411 from google:fix/agent-tool-mcp-cleanup 9ae753b5a4
PiperOrigin-RevId: 828651896

2025-11-05 15:23:39 -08:00

2.3 KiB

Raw Permalink Blame History

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.

Setup

Start the MCP simple-tool server in SSE mode (in a separate terminal):

# 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

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:

Check available tools:
```
What tools do you have access to?
```

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

Issue: #1112 - Using agent as tool outside of adk web doesn't exit cleanly
Related Issue: #929 - LiteLLM giving error with OpenAI models and Grafana's MCP server
Stdio Version: mcp_in_agent_tool_stdio - Uses local subprocess connection

2.3 KiB Raw Permalink Blame History