From d3796f9b33251d28d05e6701f11e80f02a2a49e1 Mon Sep 17 00:00:00 2001
From: Eliza Huang <heliza@google.com>
Date: Thu, 30 Oct 2025 09:38:59 -0700
Subject: [PATCH] feat: Add example for using ADK with Fast MCP sampling

Close #2893

Co-authored-by: Eliza Huang <heliza@google.com>
PiperOrigin-RevId: 826070077
---
 .../mcp_server_side_sampling/README.md        | 51 ++++++++++++
 .../mcp_server_side_sampling/__init__.py      | 15 ++++
 .../samples/mcp_server_side_sampling/agent.py | 56 +++++++++++++
 .../mcp_server_side_sampling/mcp_server.py    | 81 +++++++++++++++++++
 4 files changed, 203 insertions(+)
 create mode 100644 contributing/samples/mcp_server_side_sampling/README.md
 create mode 100755 contributing/samples/mcp_server_side_sampling/__init__.py
 create mode 100755 contributing/samples/mcp_server_side_sampling/agent.py
 create mode 100644 contributing/samples/mcp_server_side_sampling/mcp_server.py

diff --git a/contributing/samples/mcp_server_side_sampling/README.md b/contributing/samples/mcp_server_side_sampling/README.md
new file mode 100644
index 00000000..5fe96184
--- /dev/null
+++ b/contributing/samples/mcp_server_side_sampling/README.md
@@ -0,0 +1,51 @@
+# FastMCP Server-Side Sampling with ADK
+
+This project demonstrates how to use server-side sampling with a `fastmcp` server connected to an ADK `MCPToolset`.
+
+## Description
+
+The setup consists of two main components:
+
+1.  **ADK Agent (`agent.py`):** An `LlmAgent` is configured with an `MCPToolset`. This toolset connects to a local `fastmcp` server.
+2.  **FastMCP Server (`mcp_server.py`):** A `fastmcp` server that exposes a single tool, `analyze_sentiment`. This server is configured to use its own LLM for sampling, independent of the ADK agent's LLM.
+
+The flow is as follows:
+1.  The user provides a text prompt to the ADK agent.
+2.  The agent decides to use the `analyze_sentiment` tool from the `MCPToolset`.
+3.  The tool call is sent to the `mcp_server.py`.
+4.  Inside the `analyze_sentiment` tool, `ctx.sample()` is called. This delegates an LLM call to the `fastmcp` server's own sampling handler.
+5.  The `mcp_server`'s LLM processes the prompt from `ctx.sample()` and returns the result to the server.
+6.  The server processes the LLM response and returns the final sentiment to the agent.
+7.  The agent displays the result to the user.
+
+## Steps to Run
+
+### Prerequisites
+
+- Python 3.10+
+- `google-adk` library installed.
+- A configured OpenAI API key.
+
+### 1. Set up the Environment
+
+Clone the project and navigate to the directory. Make sure your `OPENAI_API_KEY` is available as an environment variable.
+
+### 2. Install Dependencies
+
+Install the required Python libraries:
+
+```bash
+pip install fastmcp openai litellm
+```
+
+### 3. Run the Example
+
+Navigate to the `samples` directory and choose this ADK agent:
+
+```bash
+adk web .
+```
+
+The agent will automatically start the FastMCP server in the background.
+
+- **Sample user prompt:** "What is the sentiment of 'I love building things with Python'?"
diff --git a/contributing/samples/mcp_server_side_sampling/__init__.py b/contributing/samples/mcp_server_side_sampling/__init__.py
new file mode 100755
index 00000000..c48963cd
--- /dev/null
+++ b/contributing/samples/mcp_server_side_sampling/__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
diff --git a/contributing/samples/mcp_server_side_sampling/agent.py b/contributing/samples/mcp_server_side_sampling/agent.py
new file mode 100755
index 00000000..36695f1b
--- /dev/null
+++ b/contributing/samples/mcp_server_side_sampling/agent.py
@@ -0,0 +1,56 @@
+# 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.
+
+import os
+
+from google.adk.agents import LlmAgent
+from google.adk.models.lite_llm import LiteLlm
+from google.adk.tools.mcp_tool import MCPToolset
+from google.adk.tools.mcp_tool.mcp_session_manager import StdioConnectionParams
+from mcp import StdioServerParameters
+
+# This example uses the OpenAI API for both the agent and the server.
+# Ensure your OPENAI_API_KEY is available as an environment variable.
+api_key = os.getenv('OPENAI_API_KEY')
+if not api_key:
+  raise ValueError('The OPENAI_API_KEY environment variable must be set.')
+
+# Configure the StdioServerParameters to start the mcp_server.py script
+# as a subprocess. The OPENAI_API_KEY is passed to the server's environment.
+server_params = StdioServerParameters(
+    command='python',
+    args=['mcp_server.py'],
+    env={'OPENAI_API_KEY': api_key},
+)
+
+# Create the ADK MCPToolset, which connects to the FastMCP server.
+# The `tool_filter` ensures that only the 'analyze_sentiment' tool is exposed
+# to the agent.
+mcp_toolset = MCPToolset(
+    connection_params=StdioConnectionParams(
+        server_params=server_params,
+    ),
+    tool_filter=['analyze_sentiment'],
+)
+
+# Define the ADK agent that uses the MCP toolset.
+root_agent = LlmAgent(
+    model=LiteLlm(model='openai/gpt-4o'),
+    name='SentimentAgent',
+    instruction=(
+        'You are an expert at analyzing text sentiment. Use the'
+        ' analyze_sentiment tool to classify user input.'
+    ),
+    tools=[mcp_toolset],
+)
diff --git a/contributing/samples/mcp_server_side_sampling/mcp_server.py b/contributing/samples/mcp_server_side_sampling/mcp_server.py
new file mode 100644
index 00000000..2680c29d
--- /dev/null
+++ b/contributing/samples/mcp_server_side_sampling/mcp_server.py
@@ -0,0 +1,81 @@
+# 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.
+
+import logging
+import os
+
+from fastmcp import Context
+from fastmcp import FastMCP
+from fastmcp.experimental.sampling.handlers.openai import OpenAISamplingHandler
+from openai import OpenAI
+
+logging.basicConfig(level=logging.INFO)
+API_KEY = os.getenv("OPENAI_API_KEY")
+
+# Set up the server's LLM handler using the OpenAI API.
+# This handler will be used for all sampling requests from tools on this server.
+llm_handler = OpenAISamplingHandler(
+    default_model="gpt-4o",
+    client=OpenAI(
+        api_key=API_KEY,
+    ),
+)
+
+
+# Create the FastMCP Server instance.
+# The `sampling_handler` is configured to use the server's own LLM.
+# `sampling_handler_behavior="always"` ensures the server never delegates
+# sampling back to the ADK agent.
+mcp = FastMCP(
+    name="SentimentAnalysis",
+    sampling_handler=llm_handler,
+    sampling_handler_behavior="always",
+)
+
+
+@mcp.tool
+async def analyze_sentiment(text: str, ctx: Context) -> dict:
+  """Analyzes sentiment by delegating to the server's own LLM."""
+  logging.info("analyze_sentiment tool called with text: %s", text)
+  prompt = f"""Analyze the sentiment of the following text as positive,
+                 negative, or neutral. Just output a single word.
+                 Text to analyze: {text}"""
+
+  # This delegates the LLM call to the server's own sampling handler,
+  # as configured in the FastMCP instance.
+  logging.info("Attempting to call ctx.sample()")
+  try:
+    response = await ctx.sample(prompt)
+    logging.info("ctx.sample() successful. Response: %s", response)
+  except Exception as e:
+    logging.error("ctx.sample() failed: %s", e, exc_info=True)
+    raise
+
+  sentiment = response.text.strip().lower()
+
+  if "positive" in sentiment:
+    result = "positive"
+  elif "negative" in sentiment:
+    result = "negative"
+  else:
+    result = "neutral"
+
+  logging.info("Sentiment analysis result: %s", result)
+  return {"text": text, "sentiment": result}
+
+
+if __name__ == "__main__":
+  print("Starting FastMCP server with tool 'analyze_sentiment'...")
+  # This runs the server process, which the ADK agent will connect to.
+  mcp.run()