I am having problem to run the microsoft MCP monitoring within a ACA

Question

I am having problem to run the microsoft MCP monitoring within a ACA

Giovanni Ramundi 0

The container keep going to fail activation with console log showing this error :

INFO: Connecting to the container 'mcp-server'...

ERROR: {"Error":{"Code":"ClusterExecEndpointWebSocketConnectionError","Message":"Failed to establish WebSocket connection with the container, error is Failed to establish WebSocket connection with Managed Environment exec endpoint, error: The server returned status code \u0027500\u0027 when status code \u0027101\u0027 was expected..","Details":null,"Target":null,"AdditionalInfo":null,"TraceId":null}}

the system log states:

{"TimeStamp":"2026-03-18 11:16:38 \u002B0000 UTC","Type":"Warning","ContainerAppName":"azure-monitor-mcp","RevisionName":"azure-monitor-mcp--0000005","ReplicaName":"azure-monitor-mcp--0000005-59d557c88c-4mqf6","Msg":"startup probe failed: connection refused","Reason":"ReplicaUnhealthy","EventSource":"ContainerAppController","Count":25}

Environment Variables are:

User's image

Rakesh Mishra 7,065 Reputation points Microsoft External Staff Moderator

2026-03-18T12:49:49.9333333+00:00
Hello Giovanni,

The 500/101 errors and the repeating startup probe failed: connection refused system logs indicate that the azure-monitor-mcp container is not passing its initial health checks.

Because the container is never officially marked as "Running" by Azure, the managed-environment exec endpoint (which relies on WebSocket) stays closed. This typically happens because default HTTP GET probes expect a 200 OK response, but the MCP endpoint expects specific JSON-RPC payloads and rejects the probe.

Here is how you can resolve this based on the official Azure Container Apps documentation:

Create a Dedicated Health Endpoint Ensure your application code has a simple GET /health endpoint that returns an HTTP 200 OK.

Update the Probe Configuration Modify your Container App's startup and liveness probes to point to /health instead of the default root path. You can do this by creating a probe-config.yaml file:
properties: template: containers: - name: azure-monitor-mcp probes: - type: startup httpGet: path: /health port: 8080 initialDelaySeconds: 5 periodSeconds: 10 - type: liveness httpGet: path: /health port: 8080 periodSeconds: 30

Apply the Configuration Run the following Azure CLI command to update your environment with the new probes:
az containerapp update \ --name azure-monitor-mcp \ --resource-group <YOUR_RG> \ --yaml probe-config.yaml
Note: If you are unable to modify the underlying container image to add a /health route, a temporary workaround is to remove the probes entirely, though keep in mind this disables automatic container restarts on failure.

Additional Check for Data API Builder (DAB) SQL MCP:

If you happen to be hosting a Data API Builder SQL MCP server rather than Azure Monitor, please also verify that you:

Use HTTP ingress (transport: auto or http).

Expose the correct target port (the default is 3000 unless overridden).

Configure CORS if connecting from VS Code or a browser-based client.

Are running Data API Builder version 1.7 or higher.

References:

Troubleshoot MCP servers on Azure Container Apps - Container fails health checks

Host MCP servers on Azure Container Apps - Ingress and networking

Quickstart: Use SQL MCP Server with Azure Container Apps

Note: This response is drafted with the help of AI systems.
Giovanni Ramundi 0 Reputation points

2026-03-18T14:47:44.0633333+00:00

@rakesh mishra this image is an official Microsoft image so I wont be able to add the new health probe. For me it is a bit shocking that the probe are not sending traffic to the right endpoint.
Rakesh Mishra 7,065 Reputation points Microsoft External Staff Moderator

2026-03-18T14:52:04.4533333+00:00

Hi Giovanni, I have reached out to you over private message regarding some information. Could you please check and share the requested information to assist you further on this.
Siddhesh Desai 4,895 Reputation points Microsoft External Staff Moderator

2026-03-26T06:38:43.7166667+00:00

Hi Giovanni Ramundi, I have sent you the bicep team to deploy mcp server with azure container app on teams, please try the same.
Praveen Kumar Gudipudi 2,275 Reputation points Microsoft External Staff Moderator

2026-04-02T14:06:03.71+00:00

Hello Giovanni Ramundi,

Could you please check the private message and provide the required details for troubleshooting the issue further.

Here is the reference link on how to access & data retention policy of private messages in Microsoft Q&A.
Praveen Kumar Gudipudi 2,275 Reputation points Microsoft External Staff Moderator

2026-04-03T15:35:56.21+00:00

Hello Giovanni Ramundi,

We haven’t heard from you on the last response and was just checking back to see if you have a resolution yet. In case if you have any resolution, please do share that same with the community as it can be helpful to others. Otherwise, will respond with more details and we will try to help.

1 answer

Your answer

Giovanni Ramundi 0 Reputation points

2026-03-18T14:47:44.0633333+00:00

@rakesh mishra this image is an official Microsoft image so I wont be able to add the new health probe. For me it is a bit shocking that the probe are not sending traffic to the right endpoint.
Rakesh Mishra 7,065 Reputation points Microsoft External Staff Moderator

2026-03-18T14:52:04.4533333+00:00

Hi Giovanni, I have reached out to you over private message regarding some information. Could you please check and share the requested information to assist you further on this.
Siddhesh Desai 4,895 Reputation points Microsoft External Staff Moderator

2026-03-26T06:38:43.7166667+00:00

Hi Giovanni Ramundi, I have sent you the bicep team to deploy mcp server with azure container app on teams, please try the same.
Praveen Kumar Gudipudi 2,275 Reputation points Microsoft External Staff Moderator

2026-04-02T14:06:03.71+00:00

Hello Giovanni Ramundi,

Could you please check the private message and provide the required details for troubleshooting the issue further.

Here is the reference link on how to access & data retention policy of private messages in Microsoft Q&A.
Praveen Kumar Gudipudi 2,275 Reputation points Microsoft External Staff Moderator

2026-04-03T15:35:56.21+00:00

Hello Giovanni Ramundi,

We haven’t heard from you on the last response and was just checking back to see if you have a resolution yet. In case if you have any resolution, please do share that same with the community as it can be helpful to others. Otherwise, will respond with more details and we will try to help.

Answer 1

The logs indicate that the container itself is failing its startup probe (connection refused), and the exec WebSocket error is a consequence of the app not being healthy rather than the root cause. Focus on getting the MCP server inside the container to start and listen correctly before using the “Connect”/exec feature.

Use the following checks and steps:

Verify the container is reachable on the configured port
- The environment variable PORT is set to 8080. Confirm that the MCP server process inside the image is actually listening on port 8080.
- In the Container Apps definition, ensure:
  - The container’s targetPort is 8080.
  - Any startup probe / liveness / readiness probes are configured to hit the correct port and path (for MCP over HTTP this is typically /mcp or whatever the server exposes).
- If the probe is configured on a different port or path, it will fail with connection refused and keep the replica in ReplicaUnhealthy state.
Confirm MCP endpoint path and transport
- With MCP_TRANSPORT=http, the server is expected to expose an HTTP JSON-RPC endpoint.
- For most MCP server frameworks on Container Apps, the common path is /mcp:
  - For .NET MapMcp → /mcp
  - For Python FastMCP → /mcp
  - For Node.js Express → /mcp
- Ensure the startup/readiness probe and any external tests use the correct path. A wrong path can result in 404s or connection issues.
Test the MCP server directly with curl
- Before using the Container Apps exec console, validate that the app responds:
```
     curl -X POST "https://<APP_NAME>.azurecontainerapps.io/mcp" \
         -H "Content-Type: application/json" \
         -d '{"jsonrpc":"2.0","id":"1","method":"initialize"}'
```
- Replace <APP_NAME> with the container app name and adjust the path if the server uses a different route.
- If this call fails or times out, the issue is with the app startup or ingress configuration, not with MCP monitoring.
Check container logs for application errors
- Use Container Apps log streaming to see why the MCP server process is exiting or not binding to the port:
```
     az containerapp logs show \
         --name azure-monitor-mcp \
         --resource-group <RESOURCE_GROUP> \
         --follow
```
- Look for stack traces, binding errors (e.g., wrong port), or configuration errors related to MCP_TRANSPORT, MCP_MODE, or workspace settings.
Validate ingress and environment
- Ensure ingress is configured correctly (for example, external ingress if the MCP server must be reachable from outside) and that the FQDN is correct.
- If using browser-based MCP clients, also ensure CORS is configured, but this is secondary to fixing the startup probe failure.

Once the container starts successfully (startup probe passes and the replica is healthy), the exec WebSocket connection from the Container Apps “Connect” experience should stop returning HTTP 500 and the MCP monitoring scenario can proceed.

References:

Troubleshoot MCP servers on Azure Container Apps

Share via

I am having problem to run the microsoft MCP monitoring within a ACA

1 answer

Your answer