MCP FAQ

​

MCP protocol

1. How is the MCP protocol implemented? MCP is an open standard protocol from Anthropic. For implementation details, see the official MCP website.

​

Deploying MCP services in the cloud

1. Why does my Firecrawl/EverArt MCP Server report an error such as "Request usage limit reached" or "Insufficient balance"? Enter the service provider's API key. Ensure that you have enough free quota or a sufficient account balance.
2. Why did my Notion/GitHub MCP Server fail to run and report an insufficient permissions error? Grant sufficient permissions to your API key. For more information, see the documentation for the specific MCP service.

​

Custom MCP services

1. Can all MCP services be deployed and used without issues? No. MCP service providers might change or shut down their services. Alibaba Cloud Model Studio only provides a way to access these services and does not guarantee their continued availability.
2. My custom MCP service deployment failed. How can I troubleshoot it? Confirm the following:
   1. The MCP service runs correctly on your local machine.
   2. The MCP service can be hosted in the cloud and does not require communication with a browser or local application.
   3. The installation method and configuration code are correct during deployment.
   4. Function Compute (FC) is activated, you have the required permissions, and your Alibaba Cloud account has no overdue payments.
3. Can I deploy an MCP service that I developed myself?
   • Node.js: Publish the package to a public npm repository and deploy it using npx <package-name>.
   • Python: Publish the package to the PyPI repository and deploy it using uvx <package-name>.
   • Deployed on a remote server: Connect using Server-Sent Events (SSE).
   For more information, see Custom MCP services.
4. Can Alibaba Cloud Model Studio MCP services be used in other MCP clients, such as Cline or Cherry Studio? Yes. You can configure MCP services within the platform, for example, in agents or workflows. You can also integrate them into third-party applications, such as Cherry Studio or Cursor, using external calls. For more information, see External calls.
5. Can Alibaba Cloud Model Studio MCP services access a local database? No. At this time, MCP services cannot access your local resources.
6. Why can't my MCP service access remote resources, such as an ApsaraDB RDS instance? MCP services are hosted on Function Compute (FC), which does not have a fixed public IP address. To allow your MCP service to access remote resources, such as an ApsaraDB RDS instance, you must configure an IP address whitelist for FC or establish a VPC connection. If your security policy does not allow you to whitelist all IP addresses (0.0.0.0), you can obtain a fixed IP address or establish a VPC connection as follows:
   1. When you create the MCP service in the Function AI console, turn on Allow VPC access in the network configuration section.
   2. Select a VPC, a vSwitch, and a security group.
   3. To use a fixed IP address, create a NAT gateway in the VPC and associate an elastic IP address (EIP) with it. Then, add that EIP to the IP address whitelist of the remote resource.
   4. To allow access only from the VPC, turn on Disable public URL and Allow only the specified VPC to call the MCP service.
7. My MCP Server is in a private npm repository. Can I deploy it to Alibaba Cloud Model Studio? No, this is not supported at this time. You must publish the server to a public npm repository or use an SSE connection instead.
8. If I deploy a custom MCP Server, can other people use it? No. Access is restricted to your Alibaba Cloud account and authorized RAM users.
9. Why can't somelocal MCP Serversbe deployed as custom services on Alibaba Cloud Model Studio? This is because they need to access local resources, such as files or hardware, that the cloud environment cannot access. You must deploy these types of services locally.
10. My MCP Server has a new version. Will the custom MCP service deployed on Alibaba Cloud Model Studio be updated automatically? No. Services deployed using npx or uvx must be manually redeployed after a version update.

​

Integration with agent or workflow applications

1. Why can't my agent application call an MCP service, or why does it report a parameter error? The agent decides which service to call and which parameters to use based on the prompt. Try the following:
   1. Optimize the prompt to clarify the intent.
   2. If that does not work, switch to a more powerful reasoning model, such as a Qwen 3 series model.
2. Why does the agent application stop responding mid-answer without an error? This issue can occur if the response length exceeds the preset limit. To resolve this issue, increase the model's Maximum Response Length.
   
3. Can MCP services be integrated when calling the Qwen API? No. Alibaba Cloud Model Studio MCP services must be integrated into an agent or workflow application. They cannot be directly integrated when you call the Qwen API.