feat: Add a load MCP resource tool

If the user specifies use_mcp_resources=True in their MCPToolset, the agent will be able to load resources with the load_mcp_resource_tool.

Co-authored-by: Kathy Wu <wukathy@google.com>
PiperOrigin-RevId: 866539602

contributing/samples/mcp_streamablehttp_agent/agent.py

@@ -52,6 +52,7 @@ Allowed directory: {_allowed_path}
                 'get_file_info',
                 'list_allowed_directories',
             ],
+            use_mcp_resources=True,
         )
     ],
 )

contributing/samples/mcp_streamablehttp_agent/filesystem_server.py

@@ -13,6 +13,7 @@
 # limitations under the License.
 
 import asyncio
+import json
 import os
 from pathlib import Path
 import sys
@@ -45,6 +46,24 @@ def get_cwd() -> str:
   return str(Path.cwd())
 
 
+# Add a resource for testing with JSON data
+@mcp.resource(
+    name="sample_data",
+    uri="file:///sample_data.json",
+    mime_type="application/json",
+)
+def sample_data() -> str:
+  data = {
+      "users": [
+          {"id": 1, "name": "Alice", "role": "admin"},
+          {"id": 2, "name": "Bob", "role": "user"},
+          {"id": 3, "name": "Charlie", "role": "user"},
+      ],
+      "settings": {"theme": "dark", "notifications": True},
+  }
+  return json.dumps(data, indent=2)
+
+
 # Graceful shutdown handler
 async def shutdown(signal, loop):
   """Cleanup tasks tied to the service's shutdown."""

src/google/adk/tools/load_mcp_resource_tool.py

@@ -0,0 +1,170 @@
+# Copyright 2026 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 __future__ import annotations
+
+import base64
+import json
+import logging
+from typing import Any
+from typing import TYPE_CHECKING
+
+from google.genai import types
+from typing_extensions import override
+
+from ..features import FeatureName
+from ..features import is_feature_enabled
+from ..models.llm_request import LlmRequest
+from .base_tool import BaseTool
+
+if TYPE_CHECKING:
+  from mcp_toolset import McpToolset
+
+  from .tool_context import ToolContext
+
+logger = logging.getLogger("google_adk." + __name__)
+
+
+class LoadMcpResourceTool(BaseTool):
+  """A tool that loads the MCP resources and adds them to the session."""
+
+  def __init__(self, mcp_toolset: McpToolset):
+    super().__init__(
+        name="load_mcp_resource",
+        description="""Loads resources from the MCP server.
+
+NOTE: Call when you need access to resources.""",
+    )
+    self._mcp_toolset = mcp_toolset
+
+  def _get_declaration(self) -> types.FunctionDeclaration | None:
+    if is_feature_enabled(FeatureName.JSON_SCHEMA_FOR_FUNC_DECL):
+      return types.FunctionDeclaration(
+          name=self.name,
+          description=self.description,
+          parameters_json_schema={
+              "type": "object",
+              "properties": {
+                  "resource_names": {
+                      "type": "array",
+                      "items": {"type": "string"},
+                  },
+              },
+          },
+      )
+    return types.FunctionDeclaration(
+        name=self.name,
+        description=self.description,
+        parameters=types.Schema(
+            type=types.Type.OBJECT,
+            properties={
+                "resource_names": types.Schema(
+                    type=types.Type.ARRAY,
+                    items=types.Schema(
+                        type=types.Type.STRING,
+                    ),
+                )
+            },
+        ),
+    )
+
+  @override
+  async def run_async(
+      self, *, args: dict[str, Any], tool_context: ToolContext
+  ) -> Any:
+    resource_names: list[str] = args.get("resource_names", [])
+    return {
+        "resource_names": resource_names,
+        "status": (
+            "resource contents temporarily inserted and removed. to access"
+            " these resources, call load_mcp_resource tool again."
+        ),
+    }
+
+  @override
+  async def process_llm_request(
+      self, *, tool_context: ToolContext, llm_request: LlmRequest
+  ) -> None:
+    await super().process_llm_request(
+        tool_context=tool_context,
+        llm_request=llm_request,
+    )
+    await self._append_resources_to_llm_request(
+        tool_context=tool_context, llm_request=llm_request
+    )
+
+  async def _append_resources_to_llm_request(
+      self, *, tool_context: ToolContext, llm_request: LlmRequest
+  ):
+    try:
+      resource_names = await self._mcp_toolset.list_resources()
+      if resource_names:
+        llm_request.append_instructions([f"""You have a list of MCP resources:
+{json.dumps(resource_names)}
+
+When the user asks questions about any of the resources, you should call the
+`load_mcp_resource` function to load the resource. Always call load_mcp_resource
+before answering questions related to the resources.
+"""])
+    except Exception as e:
+      logger.warning("Failed to list MCP resources: %s", e)
+
+    # Attach content
+    if llm_request.contents and llm_request.contents[-1].parts:
+      function_response = llm_request.contents[-1].parts[0].function_response
+      if function_response and function_response.name == self.name:
+        response = function_response.response or {}
+        resource_names = response.get("resource_names", [])
+        for resource_name in resource_names:
+          try:
+            contents = await self._mcp_toolset.read_resource(resource_name)
+
+            for content in contents:
+              part = self._mcp_content_to_part(content, resource_name)
+              llm_request.contents.append(
+                  types.Content(
+                      role="user",
+                      parts=[
+                          types.Part.from_text(
+                              text=f"Resource {resource_name} is:"
+                          ),
+                          part,
+                      ],
+                  )
+              )
+          except Exception as e:
+            logger.warning(
+                "Failed to read MCP resource '%s': %s", resource_name, e
+            )
+            continue
+
+  def _mcp_content_to_part(
+      self, content: Any, resource_name: str
+  ) -> types.Part:
+    if hasattr(content, "text") and content.text is not None:
+      return types.Part.from_text(text=content.text)
+    elif hasattr(content, "blob") and content.blob is not None:
+      try:
+        data = base64.b64decode(content.blob)
+        # Basic check for mime type or default
+        mime_type = content.mimeType or "application/octet-stream"
+        return types.Part.from_bytes(data=data, mime_type=mime_type)
+      except Exception:
+        return types.Part.from_text(
+            text=f"[Binary content for {resource_name} could not be decoded]"
+        )
+    else:
+      return types.Part.from_text(
+          text=f"[Unknown content type for {resource_name}]"
+      )

src/google/adk/tools/mcp_tool/mcp_toolset.py

@@ -16,7 +16,6 @@ from __future__ import annotations
 
 import asyncio
 import base64
-import json
 import logging
 import sys
 from typing import Any
@@ -44,8 +43,10 @@ from ...auth.auth_tool import AuthConfig
 from ..base_tool import BaseTool
 from ..base_toolset import BaseToolset
 from ..base_toolset import ToolPredicate
+from ..load_mcp_resource_tool import LoadMcpResourceTool
 from ..tool_configs import BaseToolConfig
 from ..tool_configs import ToolArgsConfig
+from ..tool_context import ToolContext
 from .mcp_session_manager import MCPSessionManager
 from .mcp_session_manager import retry_on_errors
 from .mcp_session_manager import SseConnectionParams
@@ -112,6 +113,7 @@ class McpToolset(BaseToolset):
       progress_callback: Optional[
           Union[ProgressFnT, ProgressCallbackFactory]
       ] = None,
+      use_mcp_resources: Optional[bool] = False,
   ):
     """Initializes the McpToolset.
 
@@ -148,7 +150,11 @@ class McpToolset(BaseToolset):
           progress handling logic and access/modify session state via the
           CallbackContext. The **kwargs parameter allows for future
           extensibility.
+      use_mcp_resources: Whether the agent should have access to MCP resources.
+        This will add a `load_mcp_resource` tool to the toolset and include
+        available resources in the agent context. Defaults to False.
     """
+
     super().__init__(tool_filter=tool_filter, tool_name_prefix=tool_name_prefix)
 
     if not connection_params:
@@ -177,6 +183,7 @@ class McpToolset(BaseToolset):
         if auth_scheme
         else None
     )
+    self._use_mcp_resources = use_mcp_resources
 
   def _get_auth_headers(self) -> Optional[Dict[str, str]]:
     """Build authentication headers from exchanged credential.
@@ -317,6 +324,13 @@ class McpToolset(BaseToolset):
 
       if self._is_tool_selected(mcp_tool, readonly_context):
         tools.append(mcp_tool)
+
+    if self._use_mcp_resources:
+      load_resource_tool = LoadMcpResourceTool(
+          mcp_toolset=self,
+      )
+      tools.append(load_resource_tool)
+
     return tools
 
   async def read_resource(
@@ -415,6 +429,7 @@ class McpToolset(BaseToolset):
         tool_name_prefix=mcp_toolset_config.tool_name_prefix,
         auth_scheme=mcp_toolset_config.auth_scheme,
         auth_credential=mcp_toolset_config.auth_credential,
+        use_mcp_resources=mcp_toolset_config.use_mcp_resources,
     )
 
 
@@ -451,6 +466,8 @@ class McpToolsetConfig(BaseToolConfig):
 
   auth_credential: Optional[AuthCredential] = None
 
+  use_mcp_resources: bool = False
+
   @model_validator(mode="after")
   def _check_only_one_params_field(self):
     param_fields = [

tests/unittests/tools/mcp_tool/test_mcp_toolset.py

@@ -25,6 +25,7 @@ from unittest.mock import patch
 
 from google.adk.agents.readonly_context import ReadonlyContext
 from google.adk.auth.auth_credential import AuthCredential
+from google.adk.tools.load_mcp_resource_tool import LoadMcpResourceTool
 from google.adk.tools.mcp_tool.mcp_session_manager import MCPSessionManager
 from google.adk.tools.mcp_tool.mcp_session_manager import SseConnectionParams
 from google.adk.tools.mcp_tool.mcp_session_manager import StdioConnectionParams
@@ -81,6 +82,14 @@ class TestMcpToolset:
     assert toolset._errlog == sys.stderr
     assert toolset._auth_scheme is None
     assert toolset._auth_credential is None
+    assert toolset._use_mcp_resources is False
+
+  def test_init_with_use_mcp_resources(self):
+    """Test initialization with use_mcp_resources."""
+    toolset = McpToolset(
+        connection_params=self.mock_stdio_params, use_mcp_resources=True
+    )
+    assert toolset._use_mcp_resources is True
 
   def test_init_with_stdio_connection_params(self):
     """Test initialization with StdioConnectionParams."""
@@ -161,17 +170,21 @@ class TestMcpToolset:
         return_value=MockListToolsResult(mock_tools)
     )
 
-    toolset = McpToolset(connection_params=self.mock_stdio_params)
+    toolset = McpToolset(
+        connection_params=self.mock_stdio_params, use_mcp_resources=True
+    )
     toolset._mcp_session_manager = self.mock_session_manager
 
     tools = await toolset.get_tools()
 
-    assert len(tools) == 3
-    for tool in tools:
+    assert len(tools) == 4
+    for tool in tools[:3]:
       assert isinstance(tool, MCPTool)
+    assert isinstance(tools[3], LoadMcpResourceTool)
     assert tools[0].name == "tool1"
     assert tools[1].name == "tool2"
     assert tools[2].name == "tool3"
+    assert tools[3].name == "load_mcp_resource"
 
   @pytest.mark.asyncio
   async def test_get_tools_with_list_filter(self):
@@ -338,6 +351,7 @@ class TestMcpToolset:
     toolset = McpToolset(
         connection_params=mock_connection_params,
         tool_name_prefix="my_prefix",
+        use_mcp_resources=True,
     )
 
     # Replace the internal session manager with our mock
@@ -352,13 +366,15 @@ class TestMcpToolset:
     prefixed_tools = await toolset.get_tools_with_prefix()
 
     # Assert that the tools are prefixed correctly
-    assert len(prefixed_tools) == 2
+    assert len(prefixed_tools) == 3
     assert prefixed_tools[0].name == "my_prefix_tool1"
     assert prefixed_tools[1].name == "my_prefix_tool2"
+    assert prefixed_tools[2].name == "my_prefix_load_mcp_resource"
 
     # Assert that the original tools are not modified
     assert tools[0].name == "tool1"
     assert tools[1].name == "tool2"
+    assert tools[2].name == "load_mcp_resource"
 
   def test_init_with_progress_callback(self):
     """Test initialization with progress_callback."""

tests/unittests/tools/test_load_mcp_resource_tool.py

@@ -0,0 +1,165 @@
+# Copyright 2026 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 base64
+import json
+from unittest.mock import AsyncMock
+from unittest.mock import Mock
+from unittest.mock import patch
+
+from google.adk.models.llm_request import LlmRequest
+from google.adk.tools.load_mcp_resource_tool import LoadMcpResourceTool
+from google.adk.tools.mcp_tool.mcp_toolset import McpToolset
+from google.adk.tools.tool_context import ToolContext
+from google.genai import types
+from mcp.types import BlobResourceContents
+from mcp.types import TextResourceContents
+import pytest
+
+
+class TestLoadMcpResourceTool:
+  """Test suite for LoadMcpResourceTool class."""
+
+  def setup_method(self):
+    """Set up test fixtures."""
+    self.mock_mcp_toolset = Mock(spec=McpToolset)
+    self.mock_tool_context = Mock(spec=ToolContext)
+
+  def test_init(self):
+    """Test initialization."""
+    tool = LoadMcpResourceTool(mcp_toolset=self.mock_mcp_toolset)
+    assert tool.name == "load_mcp_resource"
+    assert tool._mcp_toolset == self.mock_mcp_toolset
+
+  @pytest.mark.asyncio
+  async def test_run_async(self):
+    """Test run_async method."""
+    tool = LoadMcpResourceTool(mcp_toolset=self.mock_mcp_toolset)
+    args = {"resource_names": ["res1", "res2"]}
+    result = await tool.run_async(
+        args=args, tool_context=self.mock_tool_context
+    )
+
+    assert result["resource_names"] == ["res1", "res2"]
+    assert "temporarily inserted" in result["status"]
+
+  def test_get_declaration(self):
+    """Test _get_declaration method."""
+    tool = LoadMcpResourceTool(mcp_toolset=self.mock_mcp_toolset)
+    declaration = tool._get_declaration()
+
+    assert isinstance(declaration, types.FunctionDeclaration)
+    assert declaration.name == "load_mcp_resource"
+    # Basic schema check, precise structure depends on is_feature_enabled
+    # and implementation details which might vary.
+
+  @pytest.mark.asyncio
+  async def test_process_llm_request_injects_list(self):
+    """Test that resource list is injected when enabled."""
+    tool = LoadMcpResourceTool(mcp_toolset=self.mock_mcp_toolset)
+    llm_request = Mock(spec=LlmRequest)
+    llm_request.contents = []
+
+    # Mock list_resources
+    self.mock_mcp_toolset.list_resources = AsyncMock(
+        return_value=["res1", "res2"]
+    )
+
+    await tool.process_llm_request(
+        tool_context=self.mock_tool_context, llm_request=llm_request
+    )
+
+    llm_request.append_instructions.assert_called_once()
+    instructions = llm_request.append_instructions.call_args[0][0]
+    assert "res1" in instructions[0]
+    assert "res2" in instructions[0]
+
+  async def test_process_llm_request_loads_content_text(self):
+    """Test loading text resource content."""
+    tool = LoadMcpResourceTool(mcp_toolset=self.mock_mcp_toolset)
+    llm_request = Mock(spec=LlmRequest)
+    llm_request.contents = []
+
+    # Setup LLM request with function call response asking for "res1"
+    function_response = Mock()
+    function_response.name = "load_mcp_resource"
+    function_response.response = {"resource_names": ["res1"]}
+
+    part = Mock()
+    part.function_response = function_response
+
+    content = Mock()
+    content.parts = [part]
+    llm_request.contents = [content]
+
+    # Mock read_resource
+    text_content = TextResourceContents(
+        uri="file:///res1", mimeType="text/plain", text="hello content"
+    )
+    self.mock_mcp_toolset.read_resource = AsyncMock(return_value=[text_content])
+
+    await tool.process_llm_request(
+        tool_context=self.mock_tool_context, llm_request=llm_request
+    )
+
+    # Verify content was appended
+    assert len(llm_request.contents) == 2  # Original + new content
+    new_content = llm_request.contents[1]
+    assert new_content.role == "user"
+    assert len(new_content.parts) == 2
+    assert "Resource res1 is:" in new_content.parts[0].text
+    assert new_content.parts[1].text == "hello content"
+
+  @pytest.mark.asyncio
+  async def test_process_llm_request_loads_content_binary(self):
+    """Test loading binary resource content."""
+    tool = LoadMcpResourceTool(mcp_toolset=self.mock_mcp_toolset)
+    llm_request = Mock(spec=LlmRequest)
+    llm_request.contents = []
+
+    # Setup LLM request with function call response asking for "res1"
+    function_response = Mock()
+    function_response.name = "load_mcp_resource"
+    function_response.response = {"resource_names": ["res1"]}
+
+    part = Mock()
+    part.function_response = function_response
+
+    content = Mock()
+    content.parts = [part]
+    llm_request.contents = [content]
+
+    # Mock read_resource
+    blob_data = b"binary data"
+    blob_b64 = base64.b64encode(blob_data).decode("ascii")
+    blob_content = BlobResourceContents(
+        uri="file:///res1", mimeType="image/png", blob=blob_b64
+    )
+    self.mock_mcp_toolset.read_resource = AsyncMock(return_value=[blob_content])
+
+    await tool.process_llm_request(
+        tool_context=self.mock_tool_context, llm_request=llm_request
+    )
+
+    # Verify content was appended
+    assert len(llm_request.contents) == 2
+    new_content = llm_request.contents[1]
+    # Check that the second part is bytes
+    # Note: google.genai.types.Part.from_bytes creates a Part with inline_data
+    # Accessing it depends on the Part implementation.
+    # Since we are using real types.Part (not mocked), we can check attributes.
+    part = new_content.parts[1]
+    assert part.inline_data is not None
+    assert part.inline_data.mime_type == "image/png"
+    assert part.inline_data.data == blob_data