[BUG] mcp client doesn't support namespace prefix

[csantanapr]

Checks

• I have updated to the lastest minor and patch version of Strands
• I have checked the documentation and this is not expected behavior
• I have searched ./issues and there are no duplicates of my issue

Strands Version

1.4.0

Python Version

3.13

Operating System

macOS

Installation Method

pip

Steps to Reproduce

When creating multiple mcp clients that offer a tool with the same name, it can be multiple instances of the same mcp server package, or 2 mcp server that have generic names like execute or run Strands should offer a way to namespace with prefix in front, this is what projects like agentgateway do with mcp multiplexing

Here is an example code snippet using two instances of the EKS MCP Server

mcp_client = MCPClient(lambda: stdio_client(
    StdioServerParameters(
        command="uvx",
        args=["awslabs.eks-mcp-server@latest","--allow-write","--allow-sensitive-data-access"],
        env={
            "AWS_REGION": "us-west-2",
            "AWS_PROFILE": "cross-account-1"
        }
    )
))
mcp_client_2 = MCPClient(lambda: stdio_client(
    StdioServerParameters(
        command="uvx",
        args=["awslabs.eks-mcp-server@latest","--allow-write","--allow-sensitive-data-access"],
        env={
            "AWS_REGION": "us-west-2",
            "AWS_PROFILE": "carrlos-Admin"
        }
    )
))

Then using the mcp clients in Strands

mcp_tools = mcp_client.list_tools_sync()
mcp_tools_2 = mcp_client_2.list_tools_sync()

    # Namespace the tools to avoid conflicts
    namespaced_tools_1 = [NamespacedMCPTool(tool, "cross") for tool in mcp_tools]
    namespaced_tools_2 = [NamespacedMCPTool(tool, "carrlos") for tool in mcp_tools_2]

    all_tools = [get_todays_date, use_aws, environment, get_profile_from_cluster] + namespaced_tools_1 + namespaced_tools_2
    agent = Agent(
        name="MyAgent",
        model="us.anthropic.claude-3-7-sonnet-20250219-v1:0",
        system_prompt="""
                    You are a Platform Engineer managing EKS clusters located in different AWS accounts.

                    IMPORTANT: When working with EKS clusters, you MUST:
                    1. First call get_profile_from_cluster(cluster_name) to determine which AWS profile/namespace to use
                    2. Use the returned profile information to select the correct namespaced tools:
                    - If profile is "cross-account-1", use tools prefixed with "cross__"
                    - If profile is "carrlos-Admin", use tools prefixed with "carrlos__"

                    For example:
                    - For cluster "agentic-ai-on-eks-workshop" → use "cross__" tools
                    - For cluster "cloud-provider-1" → use "carrlos__" tools

                    Always check the cluster-to-profile mapping before executing any Kubernetes operations.
                    """,
        tools=all_tools
    )

I had to override AgentTool to inject the namespace

class NamespacedMCPTool(MCPAgentTool):
    """Wrapper for MCPAgentTool that allows custom naming"""

    def __init__(self, mcp_tool: MCPAgentTool, namespace: str):
        # Properly initialize the parent class
        super().__init__(mcp_tool.mcp_tool, mcp_tool.mcp_client)
        self.namespace = namespace
        self._custom_name = f"{namespace}__{mcp_tool.tool_name}"
        self._original_name = mcp_tool.tool_name

    @property
    def tool_name(self) -> str:
        return self._custom_name

    @property
    def tool_spec(self) -> ToolSpec:
        spec = super().tool_spec
        spec["name"] = self._custom_name
        return spec

    async def stream(self, tool_use: ToolUse, invocation_state: dict[str, Any], **kwargs: Any) -> ToolGenerator:
        """Stream the MCP tool using the original tool name for the MCP server call."""
        result = await self.mcp_client.call_tool_async(
            tool_use_id=tool_use["toolUseId"],
            name=self._original_name,  # Use original name for MCP server
            arguments=tool_use["input"],
        )
        yield result

Expected Behavior

For the mcp classes like STDIO or StreammeableHTTP to offer an optional argument name and used as prefix under the hood
then I could something like

StdioServerParameters(
        name="cross"
        command="uvx",
        args=["awslabs.eks-mcp-server@latest","--allow-write","--allow-sensitive-data-access"],
        env={
            "AWS_REGION": "us-west-2",
            "AWS_PROFILE": "cross-account-1"
        }
    )

Actual Behavior

The tools get override and a single set of tools offer to the Agent

Additional Context

Maybe related to issue #715

Possible Solution

No response

Related Issues

No response

[JackYPCOnline]

Thank you for raising this issue.
Our team is prorigtizing a new MCP interface, I am prettey sure this issue would be addressed at that time.