McpClientPlugin plugin not working - @teams/app.HttpPlugin.@teams/http-stream waiting for ID to be set

[Micatio]

Weirdly the chat send doesnt response for any MCP requests?

mcp_plugin = McpClientPlugin()
    mcp_plugin.use_mcp_server("https://learn.microsoft.com/api/mcp") 
    chat_prompt = ChatPrompt(model,
                             plugins=[mcp_plugin])

 chat_result = await chat_prompt.send(
        input=ctx.activity.text, 
        memory=memory,
        instructions=INSTRUCTIONS,
        on_chunk=lambda chunk: ctx.stream.emit(chunk)
    )

INSTRUCTIONS = "You are a helpful assistant. You MUST use tool calls to do all your work."
Sent "What are the az cli commands to create an Azure container app according to official Microsoft Learn documentation"
....No response

[Micatio]

Weirdly the chat send doesnt response for any MCP requests?

mcp_plugin = McpClientPlugin()
    mcp_plugin.use_mcp_server("https://learn.microsoft.com/api/mcp") 
    chat_prompt = ChatPrompt(model,
                             plugins=[mcp_plugin])

 chat_result = await chat_prompt.send(
        input=ctx.activity.text, 
        memory=memory,
        instructions=INSTRUCTIONS,
        on_chunk=lambda chunk: ctx.stream.emit(chunk)
    )

INSTRUCTIONS = "You are a helpful assistant. You MUST use tool calls to do all your work." Sent "What are the az cli commands to create an Azure container app according to official Microsoft Learn documentation" ....No response

Actually "list tools" does response with the below

_Here are the tools available for use:

functions.microsoft_docs_search: Searches official Microsoft/Azure documentation for relevant content.
functions.microsoft_code_sample_search: Searches for code snippets and examples in official Microsoft Learn documentation.
functions.microsoft_docs_fetch: Fetches and converts a Microsoft Learn documentation page to markdown format.
multi_tool_use.parallel: Runs multiple tools simultaneously if they can operate in parallel._

Quota issue for such a simple request?

[ERROR] @teams/app.HttpPlugin Activity c6562d95-4d5a-4c5d-9340-xxxxxxxx failed: Error code: 429 - {'error': {'code': 'RateLimitReached', 'message': 'Your requests to gpt-4o for gpt-4o in West US have exceeded the token rate limit for your current AIServices S0 pricing tier. This request was for ChatCompletions_Create under Azure OpenAI API version 2024-12-01-preview. Please retry after 60 seconds. To increase your default rate limit, visit: https://aka.ms/oai/quotaincrease.'}}
INFO: 127.0.0.1:50476 - "POST /api/messages HTTP/1.1" 500 Internal Server Error
ERROR: Exception in ASGI application

[heyitsaamir]

I tried this out, and it seems like the issue is with the learn MCP server. It seems to not be responding (for whatever reason). Can you try with a different server?
I just tried with Context7 and it seems to work as expected.

mcp_plugin = McpClientPlugin()
mcp_plugin.use_mcp_server(
    "https://mcp.context7.com/mcp",
    McpClientPluginParams(headers={"Authorization": f"Bearer {getenv('CONTEXT7_API_KEY')}"}),
)
chat_prompt = ChatPrompt(model,
                         plugins=[mcp_plugin])

chat_result = await chat_prompt.send(
      input=ctx.activity.text, 
      memory=memory,
      instructions=INSTRUCTIONS,
      on_chunk=lambda chunk: ctx.stream.emit(chunk)
  )

[Micatio]

I tried this out, and it seems like the issue is with the learn MCP server. It seems to not be responding (for whatever reason). Can you try with a different server?

I suspect the payload or whatever its sending is too large as its hitting the rate limit, how do I log the requests sent to the Open API endpoint?

[heyitsaamir]

I tried this out, and it seems like the issue is with the learn MCP server. It seems to not be responding (for whatever reason). Can you try with a different server?

I suspect the payload or whatever its sending is too large as its hitting the rate limit, how do I log the requests sent to the Open API endpoint?

You can try seeing LOG_LEVEL=debug in your environment variables.

[Micatio]

I tried this out, and it seems like the issue is with the learn MCP server. It seems to not be responding (for whatever reason). Can you try with a different server?

I suspect the payload or whatever its sending is too large as its hitting the rate limit, how do I log the requests sent to the Open API endpoint?

You can try seeing LOG_LEVEL=debug in your environment variables.

Hmmm, its actually retrieved the data correctly from learn @heyitsaamir..stuck on this line "[DEBUG] @teams/app.HttpPlugin.@teams/http-stream waiting for ID to be set"?

[DEBUG] OpenAI-Model Converted to 4 OpenAI messages
[DEBUG] OpenAI-Model Making Chat Completions API call (streaming: False)
[DEBUG] OpenAI-Model Response has 1 function calls, executing recursively
[DEBUG] OpenAI-Model Executing 1 function calls
[DEBUG] mcp_client Making call to https://learn.microsoft.com/api/mcp tool-name=microsoft_docs_fetch
[DEBUG] mcp_client Successfully received result for mcp call # Use Azure Functions in Azure Container Apps
[DEBUG] mcp_client
[DEBUG] mcp_client This article shows you how to create an Azure Functions app in Azure Container Apps, complete with preconfigured autoscaling rules.
[DEBUG] mcp_client
[DEBUG] mcp_client ## Prerequisites
[DEBUG] mcp_client
[DEBUG] mcp_client | Resource | Description |
[DEBUG] mcp_client | --- | --- |
[DEBUG] mcp_client | Azure account | An Azure account with an active subscription.If you don't have one, you can create one for free. |
[DEBUG] mcp_client | Azure Storage account | A blob storage account to store state for your Azure Functions. |
[DEBUG] mcp_client | Azure Application Insights | An instance of Azure Application Insights to collect data about your container app. |
[DEBUG] mcp_client
[DEBUG] mcp_client ## Create a Functions app
[DEBUG] mcp_client
[DEBUG] mcp_client The following steps show you how to use a sample container image to create your container app. If you want to use this procedure with a custom container image, see Create your first function app in Azure Container Apps.
[DEBUG] mcp_client
[DEBUG] mcp_client 1. Go to the Azure portal and search for Container Apps in the search bar.
[DEBUG] mcp_client 2. Select Container Apps.
[DEBUG] mcp_client 3. Select Create.
[DEBUG] mcp_client 4. Select Container App
[DEBUG] mcp_client 5. In the Basics section, enter the following values.
[DEBUG] mcp_client
[DEBUG] mcp_client Under Project details:
[DEBUG] mcp_client
[DEBUG] mcp_client | Property | Value |
[DEBUG] mcp_client | --- | --- |
[DEBUG] mcp_client | Subscription | Select your Azure subscription. |
[DEBUG] mcp_client | Resource group | Select Create new resource group, name it my-aca-functions-group, and select OK. |
[DEBUG] mcp_client | Container app name | Enter my-aca-functions-app. |
[DEBUG] mcp_client 6. Next to Optimize for Azure Functions, check the checkbox.
[DEBUG] mcp_client 7. Under Container Apps environment, enter:
[DEBUG] mcp_client
[DEBUG] mcp_client | Property | Value |
[DEBUG] mcp_client | --- | --- |
[DEBUG] mcp_client | Region | Select a region closest to you. |
[DEBUG] mcp_client | Container Apps environment | Select Create new environment. |
[DEBUG] mcp_client 8. In the environment setup window, enter:
[DEBUG] mcp_client
[DEBUG] mcp_client | Property | Value |
[DEBUG] mcp_client | --- | --- |
[DEBUG] mcp_client | Environment name | Enter my-aca-functions-environment |
[DEBUG] mcp_client | Zone redundancy | Select Disabled. |
[DEBUG] mcp_client 9. Select Create to save your values.
[DEBUG] mcp_client 10. Select Next: Container to switch to the Container section.
[DEBUG] mcp_client 11. Next to Use quickstart image, leave this box unchecked.
[DEBUG] mcp_client 12. Under the Container details section, enter the following values.
[DEBUG] mcp_client
[DEBUG] mcp_client | Property | Value |
[DEBUG] mcp_client | --- | --- |
[DEBUG] mcp_client | Name | This box is prefilled with your selection in the last section. |
[DEBUG] mcp_client | Image source | Select Docker Hub or other registries |
[DEBUG] mcp_client | Subscription | Select your subscription. |
[DEBUG] mcp_client | Image type | Select Public. |
[DEBUG] mcp_client | Registry login server | Enter mcr.microsoft.com |
[DEBUG] mcp_client | Image and tag | Enter k8se/quickstart-functions:latest |
[DEBUG] mcp_client 13. Under Environment variables, enter values for the following variables:
[DEBUG] mcp_client
[DEBUG] mcp_client - AzureWebJobsStorage
[DEBUG] mcp_client - APPINSIGHTS_INSTRUMENTATIONKEY or APPLICATIONINSIGHTS_CONNECTION_STRING
[DEBUG] mcp_client
[DEBUG] mcp_client Enter either Managed identity or connection string values for these variables. Managed Identity is recommended.
[DEBUG] mcp_client
[DEBUG] mcp_client The AzureWebJobsStorage variable is a required Azure Storage account connection string for Azure Functions. This storage account stores function execution logs, manage triggers and bindings, and maintains state for durable functions.
[DEBUG] mcp_client
[DEBUG] mcp_client Application Insights is a monitoring and diagnostic service that provides insights into the performance and usage of your Azure Functions. This monitoring helps you track request rates, response times, failure rates, and other metrics.
[DEBUG] mcp_client 14. Select Next > Ingress to switch to the Ingress section and enter the following values.
[DEBUG] mcp_client
[DEBUG] mcp_client | Property | Value |
[DEBUG] mcp_client | --- | --- |
[DEBUG] mcp_client | Ingress | Select the Enabled checkbox to enable ingress. |
[DEBUG] mcp_client | Ingress traffic | Select Accepting traffic from anywhere. |
[DEBUG] mcp_client | Ingress type | Select HTTP. |
[DEBUG] mcp_client | Target port | Enter 80. |
[DEBUG] mcp_client 15. Select Review + Create.
[DEBUG] mcp_client 16. Select Create.
[DEBUG] mcp_client 17. Once the deployment is complete, select Go to resource.
[DEBUG] mcp_client 18. From the Overview page, select the link next to Application URL to open the application in a new browser tab.
[DEBUG] mcp_client 19. Append /api/HttpExample to the end of the URL.
[DEBUG] mcp_client
[DEBUG] mcp_client A message stating "HTTP trigger function processed a request" is returned in the browser.
[DEBUG] mcp_client
[DEBUG] mcp_client ## Prerequisites
[DEBUG] mcp_client
[DEBUG] mcp_client - An Azure account with an active subscription.
[DEBUG] mcp_client - If you don't have one, you can create one for free.
[DEBUG] mcp_client - Install the Azure CLI.
[DEBUG] mcp_client
[DEBUG] mcp_client ## Create a Functions App
[DEBUG] mcp_client
[DEBUG] mcp_client To sign in to Azure from the CLI, run the following command and follow the prompts to complete the authentication process.
[DEBUG] mcp_client
[DEBUG] mcp_client 1. Sign in to Azure.
[DEBUG] mcp_client
[DEBUG] mcp_client azurecli [DEBUG] mcp_client az login [DEBUG] mcp_client
[DEBUG] mcp_client 2. To ensure you're running the latest version of the CLI, run the upgrade command.
[DEBUG] mcp_client
[DEBUG] mcp_client azurecli [DEBUG] mcp_client az upgrade [DEBUG] mcp_client
[DEBUG] mcp_client 3. Install or update the Azure Container Apps extension for the CLI.
[DEBUG] mcp_client
[DEBUG] mcp_client If you receive errors about missing parameters when you run az containerapp commands in Azure CLI or cmdlets from the Az.App module in PowerShell, be sure you have the latest version of the Azure Container Apps extension installed.
[DEBUG] mcp_client
[DEBUG] mcp_client azurecli [DEBUG] mcp_client az extension add --name containerapp --upgrade [DEBUG] mcp_client
[DEBUG] mcp_client
[DEBUG] mcp_client Now that the current extension or module is installed, register the Microsoft.App and Microsoft.OperationalInsights namespaces.
[DEBUG] mcp_client
[DEBUG] mcp_client azurecli [DEBUG] mcp_client az provider register --namespace Microsoft.App [DEBUG] mcp_client
[DEBUG] mcp_client
[DEBUG] mcp_client azurecli [DEBUG] mcp_client az provider register --namespace Microsoft.OperationalInsights [DEBUG] mcp_client
[DEBUG] mcp_client 4. Create environment variables.
[DEBUG] mcp_client
[DEBUG] mcp_client bash [DEBUG] mcp_client RESOURCE_GROUP_NAME="my-aca-functions-group" [DEBUG] mcp_client CONTAINER_APP_NAME="my-aca-functions-app" [DEBUG] mcp_client ENVIRONMENT_NAME="my-aca-functions-environment" [DEBUG] mcp_client LOCATION="westus" [DEBUG] mcp_client
[DEBUG] mcp_client 5. Create a resource group.
[DEBUG] mcp_client
[DEBUG] mcp_client azurecli [DEBUG] mcp_client az group create \ [DEBUG] mcp_client --name $RESOURCE_GROUP_NAME \ [DEBUG] mcp_client --location $LOCATION \ [DEBUG] mcp_client --output none [DEBUG] mcp_client
[DEBUG] mcp_client 6. Create the Container Apps environment.
[DEBUG] mcp_client
[DEBUG] mcp_client azurecli [DEBUG] mcp_client az containerapp env create \ [DEBUG] mcp_client --name $ENVIRONMENT_NAME \ [DEBUG] mcp_client --resource-group $RESOURCE_GROUP_NAME \ [DEBUG] mcp_client --location $LOCATION \ [DEBUG] mcp_client --output none [DEBUG] mcp_client
[DEBUG] mcp_client 7. Create an Azure Functions container app.
[DEBUG] mcp_client
[DEBUG] mcp_client azurecli [DEBUG] mcp_client az containerapp create \ [DEBUG] mcp_client --resource-group $RESOURCE_GROUP_NAME \ [DEBUG] mcp_client --name $CONTAINER_APP_NAME \ [DEBUG] mcp_client --environment $ENVIRONMENT_NAME \ [DEBUG] mcp_client --image mcr.microsoft.com/azure-functions/dotnet8-quickstart-demo:1.0 \ [DEBUG] mcp_client --ingress external \ [DEBUG] mcp_client --target-port 80 \ [DEBUG] mcp_client --kind functionapp \ [DEBUG] mcp_client --query properties.outputs.fqdn [DEBUG] mcp_client
[DEBUG] mcp_client
[DEBUG] mcp_client This command returns the URL of your Functions app. Copy this URL and paste it into a web browser.
[DEBUG] mcp_client 8. Append /api/HttpExample to the end of the URL.
[DEBUG] mcp_client
[DEBUG] mcp_client A message stating "HTTP trigger function processed a request" is returned in the browser.
[DEBUG] OpenAI-Model Retrieved 3 messages from memory, 1 function results
[DEBUG] OpenAI-Model Converted to 6 OpenAI messages
[DEBUG] OpenAI-Model Making Chat Completions API call (streaming: False)
[DEBUG] OpenAI-Model Chat Completions conversation completed, returning response
[DEBUG] @teams/app.HttpPlugin.@teams/http-stream waiting for ID to be set

[Micatio]

Its blocked waiting for some other event given the mcp_client response is successful

http_stream.py
_# Wait until _id is set and queue is empty
if not self._id:
self._logger.debug("waiting for ID to be set")
await self.id_set_event.wait()

The events are set before receiving the mcp_client payload

[heyitsaamir]

@Micatio aha, it's because of on_chunk=lambda chunk: ctx.stream.emit(chunk). Currently MCP Plugin doesn't do any streaming. Does your use case require streaming?

[Micatio]

@Micatio aha, it's because of on_chunk=lambda chunk: ctx.stream.emit(chunk). Currently MCP Plugin doesn't do any streaming. Does your use case require streaming?

Nope, I can remove on_chunk completely from send however that causes "list tools" message to fail at the same position, whats the correct removely of on_chunk?

_flush(self) method doesnt handle the response correctly at the moment.

[heyitsaamir]

just remove it? or pass it as on_chunk=None?

[Micatio]

just remove it? or pass it as on_chunk=None?

Doesnt work, maybe because its mixed with the OpenAICompletionsAIModel model.

Its still referencing the http-stream class.

[heyitsaamir]

Digging into it

[Micatio]

Digging into it

Good luck walking down that rabbit hole :-)

[heyitsaamir]

Hm, today the learn docs seem more responsive and actually working. Just tried this and it worked. Can you try the same @Micatio ?

"""
Copyright (c) Microsoft Corporation. All rights reserved.
Licensed under the MIT License.
"""

import asyncio
from os import getenv

from dotenv import find_dotenv, load_dotenv
from microsoft.teams.ai import ChatPrompt
from microsoft.teams.api import MessageActivity, MessageActivityInput, TypingActivityInput
from microsoft.teams.apps import ActivityContext, App
from microsoft.teams.devtools import DevToolsPlugin
from microsoft.teams.mcpplugin import McpClientPlugin
from microsoft.teams.openai import OpenAICompletionsAIModel

load_dotenv(find_dotenv(usecwd=True))

app = App(plugins=[DevToolsPlugin()])


def get_required_env(key: str) -> str:
    value = getenv(key)
    if not value:
        raise ValueError(f"Required environment variable {key} is not set")
    return value


AZURE_OPENAI_MODEL = get_required_env("AZURE_OPENAI_MODEL")


@app.on_message
async def handle_fallback_message(ctx: ActivityContext[MessageActivity]):
    """Fallback handler using Agent with MCP tools"""
    await ctx.send(TypingActivityInput())

    mcp_plugin = McpClientPlugin()
    mcp_plugin.use_mcp_server("https://learn.microsoft.com/api/mcp")
    chat_prompt = ChatPrompt(OpenAICompletionsAIModel(model=AZURE_OPENAI_MODEL), plugins=[mcp_plugin])
    result = await chat_prompt.send(ctx.activity.text)
    if result.response.content:
        message = MessageActivityInput(text=result.response.content).add_ai_generated()
        await ctx.send(message)


if __name__ == "__main__":
    asyncio.run(app.start())

Could you share your full including:

1. Version numbers
2. OS?