MCP Hub

Yandex AI Studio enables using MCP servers in AI agents to work with external systems to get additional context or perform actions, e.g., by calling an external API.

MCP Hub is a AI Studio tool for creating, managing, and monitoring MCP servers. MCP Hub allows you to hook up external MCP servers or create your own either from scratch or using pre-configured templates.

To access MCP servers in MCP Hub, you need the serverless.mcpGateways.invoker role or higher. To access external MCP servers and MCP servers created from a template, you will additionally need the serverless.mcpGateways.anonymousInvoker role or higher.

MCP serversMCP servers

MCP servers enable large language models (LLMs) to integrate with third-party tools, such as APIs, databases, and various services that provide access to external information and allow handling complex tasks. MCP servers use the MCP (Model Context Protocol) to ensure model communication with those third-party tools.

The MCP enables you to:

• Integrate LLMs into enterprise systems, services, and internal knowledge bases.
• Provide models with standardized access to external data and tools.
• Streamline the development and scaling of LLM-based solutions.

MCP transport methodsMCP transport methods

The MCP supports communication through one of the following transport methods:

• HTTP Stream is the latest state-of-art transport mechanism. Your AI agent will be able to get updates (events) from the server without a persistent HTTP connection.
• SSE is a legacy transport mechanism. Your AI agent will be able to get updates from the server over the same persistent HTTP connection.

MCP server use casesMCP server use cases

MCP servers support the following use cases:

• Connecting models to CRM and ERP systems or internal storages.
• Using AI to automate business and operational workflows.
• Building field-specific AI assistants (legal, medical, technical, and more).
• Integrating with external APIs to access up-to-date information, such as news or exchange rates.

Adding MCP servers in MCP HubAdding MCP servers in MCP Hub

In MCP Hub, you can add MCP servers in three ways: connect an existing external server, build a new server from a template, or build a new one from scratch.

To create an MCP server, you need the serverless.mcpGateways.editor role or higher.

Connecting an external MCP serverConnecting an external MCP server

This approach works well if you have already deployed an external MCP server, e.g., on smithery.ai, from Yandex Cloud Marketplace, or on a Yandex Compute Cloud VM. To add such an MCP server to MCP Hub, provide the server address and authentication credentials. This will make it available for connection to agents.

For more information, see Connecting an external MCP server to MCP Hub.

Instead of MCP Hub, you can use the Responses API to access an external MCP server from an AI agent. To do this, simply specify the server URL and the API key.

from openai import OpenAI

# Creating a client
client = openai.OpenAI(
    api_key="<API_key>",
    base_url="https://rest-assistant.api.cloud.yandex.net/v1",
    project="<folder_ID>"
)
# Example of calling the MCP via the Responses API
response = client.responses.create(
    model="gpt://<folder_ID>/yandexgpt",
    input=[
        {
            "role": "user",
            "content": "Find a client named John Doe in your CRM"
        }
    ],
    # MCP tools, available models
    tools=[
        {
            "server_label": "crm_lookup",
            "server_url": "<MCP_server_URL>",
            "type": "mcp",
            "metadata": {
                "description": "Client search in CRM by name"
            }
        }
    ]
)

print(response.output_text)

Building an MCP server from a templateBuilding an MCP server from a template

AI Studio offers some MCP server templates from external Yandex Cloud partners. You can use these in your AI agents to access these partners' services. To add such an MCP server to MCP Hub, select the template you need and provide authentication credentials. This will make the server available for connection to your agents.

For more information, see Building an MCP server from a template in MCP Hub.

For the list of available templates, see MCP server templates.

Building an MCP server from scratchBuilding an MCP server from scratch

In MCP Hub, you can manually build a new MCP server featuring the following tools:

• Function in Yandex Cloud Functions: Enables you to run your code in one of the supported programming languages with the required dependencies within a managed runtime environment. When you increase the number of function calls, Yandex Cloud Functions performs autoscaling and creates additional instances of your function.
• HTTP request: Enables you to access external APIs. If your service already exposes an API, you can adapt it for MCP. Describe the available API methods, and they will be automatically converted for use in agents.
• Workflow in Yandex Workflows: Enables you to create complex workflows composed of interacting tools and services that implement their own control and error-handling logic and integrate these workflows with the agent.

For more information, see Creating an MCP server in MCP Hub from scratch.

General MCP server settingsGeneral MCP server settings

MCP servers in MCP Hub can be either private or public.

Unauthenticated access to a private MCP server in a folder is only allowed for AI agents initiated by the user or service account assigned a role permitting access to MCP servers in this folder. Other agents must authenticate with a service account's IAM token or API key to connect to a private MCP server.

Public MCP servers allow unauthenticated access for all agents.

In the MCP server settings, you can enable logging to save server access information to a log.

See alsoSee also

• MCP server templates
• Connecting an external MCP server to MCP Hub
• Building an MCP server from a template in MCP Hub
• Creating an MCP server in MCP Hub from scratch