The COROS MCP connects your COROS account to AI tools like Claude and ChatGPT, so you can ask direct questions about your training, build your own tools or dashboards, and gain deeper insights about your performance data. This guide covers what the MCP is, what it can and cannot do, and how to set it up.
What is an MCP?
MCP stands for Model Context Protocol. It lets an AI model securely connect to an outside data source - in this case, your COROS training history - so the AI can read that data and answer questions about it directly.
With the COROS MCP connected, AI products like Claude and ChatGPT can access your actual numbers and respond accordingly. Ask something like:
- "How did my running look this month?"
- "I have a race in six weeks. Am I ready?"
- "How has my resting heart rate changed since I started altitude training?"
Requirements
Here's what you need to start using the COROS MCP:
- A COROS account. If you use the COROS app, you already have one.
- A paid AI subscription to ChatGPT Plus or Claude Pro.
For best results, we recommend using Claude Pro. ChatGPT only supports MCP connections in Developer Mode, which disables memory.
While Gemini currently does not support custom MCP connectors through its standard chat interface, you can still connect via the Gemini CLI. Follow Google's setup guide here and use your region's COROS MCP link from Step 1 below as the server URL.
How to Set Up
- Copy the COROS MCP link: https://mcp.coros.com/mcp.
-
Connect it to your AI tool:
ChatGPT
1. Open ChatGPT, go to Settings → Apps → Advanced Settings
2. Turn on Developer Mode
3. Select Add App
4. Paste the link from Step 1
5. Authorize and log into your COROS account
Claude (Desktop app)
1. Open Claude and select the gear icon for Settings
2. Go to Connectors → Add Custom Connector
3. Paste the link from Step 1
4. Log into your COROS account when prompted
5. Configure the connector to allow your data to flow through
6. Select Save
The interface may look slightly different depending on your region or plan, but the steps are the same. - Confirm the connection by asking a question about your COROS data to your AI. For example, type: "Show me my workouts from the past two weeks" and if it returns your actual data, the connection is live.
What the MCP Can See
As of June 2026 update, the COROS MCP can query the following data:
-
Activity data
- Activity records
- Lap and segment data for workouts
- Workout feedback and training notes
- .FIT files, including GPS tracks (capped at 50 file requests per day)
-
Health and daily data
- Menstrual cycle data
- Daily steps, calories, workout duration, and weekly Training Load
- Sleep data
- Daily heart rate, HRV, stress, Wellness Checks, and resting heart rate
-
Fitness data
- Current training plan and training calendar
- VO2 Max, Running Fitness, race prediction times, and recovery status
- Threshold pace, and pace zones
- User profile and connected device information
Some example prompts include:
- "Build me a dashboard that compares my multi-sport training from 2025 to 2026."
- "How has my HRV responded around hard training blocks this month?"
- "Create a graph that compares my Sleep and HRV on weekdays compared to weekends."
- "I have a half marathon in five weeks. Based on my recent training, what should the next few weeks look like?"
For deeper workout analysis, you will need to request .FIT files through the MCP. The following analyses are not available in a basic query, but require pulling the .FIT files first so the MCP has the raw data to analyze:
- GPS information: The AI cannot describe a route, compare elevation profiles between two runs on the same loop, or analyze terrain from a basic query alone.
- Interval and lap analysis: While the AI can reference your lap summary for a workout, it won't be able to analyze split-by-split details if you are trying to compare across workouts or identify trends.
- Cycling power: Power data is not part of the standard query set, even when a power meter is paired and recording. This means a power curve and FTP estimation is not possible.
.FIT file requests via the COROS MCP are capped at 50 files per calendar day, but you can manually export specific .FIT files (or your entire activity history) from t.coros.com and upload the files to your AI as a workaround. See our guide here on how to export your historical activity data.
Troubleshooting
Confirm your COROS account works normally at t.coros.com first. If you are unable to log in there, the MCP connection will not work either. Try using the password reset option at t.coros.com to reset your COROS credentials. If that doesn't work, double check that you are logging in with the correct email address by going to your COROS app > Profile page > Settings > Account Management > Email.
This usually means your AI platform requires a paid plan for this feature. Free tiers often don't support MCP connections. We recommend using Claude Pro or ChatGPT plus.
ChatGPT Plus, Claude Pro, and Cursor all support MCP connections. While Gemini's standard chat interface does not currently support custom MCP connectors, the Gemini CLI does. Use the COROS MCP link from Step 1 above as the server URL if setting up with Gemini CLI.
This generally points to a temporary service issue on the API side rather than your setup. Try again after a few minutes. If it persists, confirm your connector permissions are still active in your AI platform's settings.
Yes. Once connected, the MCP queries data across every device registered to your account, not just the one most recently synced.
No. The MCP only queries your own activity data. It cannot read data belonging to another athlete or to an athlete you coach. The same permission structure that governs coach and athlete visibility in Training Hub applies here. You can manage your athlete and coach settings by going to t.coros.com, selecting your profile picture near the bottom left corner, and scrolling down to Privacy Settings.
Access to your COROS data is read-only currently. The MCP cannot write back to your account or modify your upcoming plans at this time. Your AI tool can still build a training plan or workout for you using that data as reference, but the training plan or workout remains in Claude/ChatGPT. We plan to release an update before September 2026 that will allow you to write training plans and workouts directly into the COROS app via AI.
Yes, most platforms allow you to customize which tools are accessible. Customization varies by platform. For example, in Claude Pro, go to Settings → Connectors → COROS MCP from the desktop app, and here you can specify which individual tools have full access, are approval-required, or blocked.
The MCP does not introduce new privacy risk beyond what already exists when using your COROS account and AI platform independently. The MCP is not a backdoor, nor does it bypass either platform's existing authentication or privacy controls. Data retention after a query is governed by the AI platform's own policies. See our COROS Privacy Policy and your AI platform's security documentation for specifics.
Privacy and Security
Using the COROS MCP does not create any new security or privacy risks beyond those that already exist when using your COROS account and your chosen AI platform separately. It is a structured connection layer that allows the two systems to communicate in a controlled, permission-based way. The same standards of account security, data protection, and platform privacy continue to apply.
Your COROS data remains governed by COROS privacy and security protections, and your AI interactions remain governed by the policies, permissions, and settings of the AI platform you choose to use. In other words, connecting through MCP does not change the underlying rules around how your data is protected - it simply allows approved data and actions to move between systems in a more useful and efficient way.
Access is only granted through explicit user authorization. You decide whether to connect your account, and the connection only operates within the permissions you approve. Depending on the capabilities enabled, that may include both read access to analyze training data and write access for supported actions within the connected experience. Those permissions are not open-ended; they are limited to the scope of the integration and the controls provided by COROS and the AI platform.
Importantly, this integration is not a backdoor into your account, and it does not bypass existing authentication, privacy controls, or security safeguards. It uses the same identity, authorization, and protection principles that already apply across secure connected services.
In practical terms, users should think of this as a secure extension of the tools they already use: your data protections remain in place, your permissions still matter, and you remain in control of what is connected and what actions are allowed.