Model Context Protocol (MCP) is an open protocol recently introduced by Anthropic that standardizes how AI applications provide context to LLMs. Like USB-C provides a standard way to connect devices, MCP provides a standard way for AI models to connect to various data sources and tools, and extend the capability of AI models.
ConsoleX AI provides comprehensive support for MCP tools, including both SSE-based and Stdio-based MCP servers.
How to Integrate and Use MCP Servers in ConsoleX
Adding and using MCP services in ConsoleX AI is straightforward. Just follow these steps:
Log in to ConsoleX AI, go to the "Tools" panel, switch to the "MCP server" tab, enter the JSON description of the MCP server, and click the "Save" button to complete the integration.
In auto tool use mode, add commonly used MCP servers to the auto-call list, or in manual mode, add the MCP server to the current conversation.
When starting a new chat, the AI model will autonomously decide whether to call the corresponding MCP tools based on the conversation context.
How to Configure SSE-based MCP Servers
Adding an SSE-based MCP server in ConsoleX AI is simple. Just describe the MCP server by JSON schema in the following format:
{
"url": "<your_sse_mcp_server_endpoint>"
}
You can add additional parameters, which will be passed to the server through header attributes. For example, you can add header authentication or pass other configuration parameters.
{
"url": "<your_sse_mcp_server_endpoint>",
"Authorization": "Bearer <your_auth_token>",
"timezone": "Europe/Berlin"
}
The SSE server needs to extract parameters from the Header for authentication or use in functional logic.
How to Configure Stdio-based MCP Servers
ConsoleX is a cloud service, so it cannot directly call local resources through Stdio, but you can use MCP bridge for relay to implement Stdio transmission protocol MCP server calls.
Configuration steps:
Install and run MCP Connect on your local computer and set up auth_token
Make your MCP bridge URL publicly accessible through a Tunnel. MCP bridge has built-in ngrok support, but you can also use other methods like Cloudflare Zero Trust
Enter the global MCP proxy URL and auth_token in ConsoleX's Settings, then you can add Stdio-based MCP server configurations just like a local MCP client
Besides setting global proxy endpoint and auth token, you can also set separate bridge_url and auth_token for each MCP server. Individually set bridge_url and auth_token will override the global configuration.
For example:
{
"bridge_url": "<your_mcp_bridge_url>",
"auth_token": "<your_auth_token>",
"command": "uvx",
"args": ["mcp-server-fetch"]
}
FAQ
Q: Which models support MCP servers?
A: The MCP protocol was initiated by Anthropic, so Claude 3.5 model series provides better support for MCP server calls. However, the tool portion of MCP servers is available to any large language model that supports tool calls.
Q: What are the limitations of using MCP servers in ConsoleX?
A: In ConsoleX, you cannot directly add Stdio-based MCP servers, but you can implement Stdio transmission protocol MCP calls through MCP bridge.
Q: What should I do if I can't successfully add a local Stdio-based MCP server?
A: First, make sure MCP bridge is successfully installed and running, and the MCP bridge endpoint is publicly accessible. This can be verified through the health check endpoint.
If the MCP bridge endpoint is publicly accessible but you cannot successfully add or call the MCP server, check the MCP bridge output logs in the terminal for further investigation.