The Model Context Protocol (MCP) is an open standard that enables AI assistants to connect to external tools and data sources.
With jupyterlite-ai, you can connect to remote MCP servers to extend the AI’s capabilities with additional tools.
Configuring an MCP Server¶
In JupyterLab, open the AI settings panel
Go to the MCP Servers tab
Click on Add Server
Enter the following details:
Server Name: A friendly name to identify the server
Server URL: The URL of the MCP server
Enable the server using the toggle switch
Once connected, the MCP server’s tools will be available to the AI assistant in the chat.
Example MCP Servers¶
Here are some publicly available remote MCP servers you can use with jupyterlite-ai:
DeepWiki¶
DeepWiki provides AI-generated documentation for open-source repositories.
Server URL:
https://mcp.deepwiki.com/mcp
The DeepWiki MCP server allows the AI to search and retrieve documentation about open-source projects.
GitMCP¶
GitMCP allows you to use any GitHub project as context for your conversations.
Server URL:
https://gitmcp.io/docs
The GitMCP server enables the AI to fetch and understand code from GitHub repositories.
Finding More MCP Servers¶
The MCP ecosystem is growing rapidly, with directories of available servers at places like the MCP Servers Repository.
However, be aware that most MCP servers listed in these directories will not work with jupyterlite-ai due to browser constraints explained below.
Browser Constraints¶
Since jupyterlite-ai runs entirely in the browser, it can only connect to MCP servers that meet two requirements:
Streamable HTTP transport: The server must support the Streamable HTTP transport defined in the MCP specification. The vast majority of MCP servers are designed to run locally using the stdio transport, which is inaccessible from a browser.
CORS headers: The server must have proper CORS (Cross-Origin Resource Sharing) headers configured to allow requests from web applications. Without these headers, the browser will block the connection.
In practice, this means most MCP servers will simply not work. Only servers specifically designed for browser-based clients—like the DeepWiki and GitMCP examples above—will be compatible.