The Model Context Protocol (MCP) is an open protocol that standardises communication between large language models (LLMs) and external systems. It enables AI models to access specific tools and data sources to solve complex tasks and generate domain-specific results.

Basic concepts of the MCP

MCP Server

An MCP server provides specific tools and data sources that can be used by AI models and can be compared, in a figurative sense, to the endpoints of an HTTP API that have an extended description. Examples of MCP servers are:

• Tredict MCP Server - Provides access to endurance sports training data, capacity values and training planning
• Weather data server - Provides current and historical weather information
• Calendar server - Enables the management of appointments and events

MCP Host

An MCP host is the parent application (e.g. Claude.ai in the browser or Ollama for local models) that embeds the AI model, manages the MCP client and acts as an intermediary between the user and the connected MCP servers. In addition, an MCP host can display and execute interactive MCP apps directly in the chat history, allowing users to use them without changing platforms. Examples of MCP hosts are:

• Claude Code - An AI-powered development tool from Anthropic
• Mistral Le Chat – An AI platform from France
• ChatGPT – An MCP host from OpenAI

Examples of MCP hosts with support for interactive UI elements:

• Claude.ai - Anthropic's web platform that can display MCP apps such as training plan views directly in chat
• ChatGPT – OpenAI's web interface
• Visual Studio Code

These platforms can run special MCP apps such as show-activity-ui (Activity details) or show-plan-ui (Training plan view) from Tredict directly into the chat history. This enables seamless integration of interactive applications into the AI conversation.

MCP Client

An MCP client is an application or component that communicates with MCP servers on behalf of an AI model or MCP host in order to call tools, query data and execute actions.

MCP Tools

MCP tools are specific functions provided by an MCP server. Each tool has a clearly defined interface with parameters and return values that are primarily described for evaluation by an LLM. Examples of MCP tools in the Tredict context are:

• activity-list - Lists executed training sessions
• capacity - Displays capacity values such as FTP, FTPa and HRmax
• plan-creation - Creates individual training plans
• zones - Manages intensity zones for different sports

How does the communication work?

Communication between the MCP client and MCP server follows roughly this sequence:

1. The MCP client authenticates itself with the MCP server (e.g. via API key or OAuth).
2. The client queries the available tools and their interfaces.
3. The user or an agent prompts the LLM.
4. The client now calls specific MCP tools with the required parameters as needed.
5. The MCP server executes the request and returns structured data.
6. The client processes the response and can call up additional tools.

Usage with Tredict

With the Tredict MCP Server, you can use your training data directly in your preferred AI environment:

1. Connect your MCP client (e.g. Claude Code, Mistral Vibe or ChatGPT) to the Tredict MCP server.
2. Authenticate yourself using OAuth or your personal API key, which you can create at Settings -> Personal API / MPC -> Access token for your own use.
3. Use the available tools to analyse training data, determine capacities or create training plans.
4. Receive tailored recommendations based on your individual data.

It is possible to connect any generic MCP client that supports bearer token authentication to the Tredict MCP server.
For detailed connection instructions for Claude Web, Claude Code, OpenAI Codex, OpenAI ChatGPT, Mistral Le Chat or Mistral Vibe, please refer to this link: FAQ entries on AI integrations

A detailed description of all available tools, the ‘MCP Server’ address and how to use them can be found in the Tredict MCP Server Documentation.