COROS MCP Explained: Connect Training Data to ChatGPT and Claude
COROS MCP is a no-code connector that lets a compatible AI assistant answer questions using training and health data you authorize from your COROS account. COROS currently documents setup for ChatGPT and Claude, and also lists Cursor as compatible. COROS recommends a paid ChatGPT Plus or Claude Pro account because free accounts can restrict custom connectors or reach usage limits quickly, but its current guide does not describe paid access as an absolute requirement. The connection is useful for reviewing patterns and preparing questions; it does not turn a general AI assistant into a certified coach or medical professional.
The most important current limitation is simple: the consumer MCP connection is read-only. It can retrieve authorized COROS data, but it cannot currently change your COROS calendar or write a training plan back to your account. That is different from COROS’s separate developer route, which can support training-plan writes.
What COROS MCP does
MCP stands for Model Context Protocol. Instead of exporting files and pasting tables into a chat, you connect the official COROS MCP endpoint to a supported assistant and authorize access to your account. The assistant can then request specific COROS data when it needs that context for your question.
COROS documents access to several useful groups of data:
- activities, laps, segments, notes and workout feedback;
- daily steps, calories, workout duration and weekly training load;
- sleep, heart rate, HRV, stress, wellness and resting heart rate;
- current plans and calendar entries;
- VO2 max, Running Fitness, race predictions, recovery and threshold zones;
- profile and device information;
- FIT files and GPS tracks, subject to a limit of 50 file requests per day.
Some detailed analysis requires the original FIT file. COROS specifically notes that deeper GPS and split-by-split analysis use these files, while cycling power is not available through the standard activity query.
COROS MCP vs the regular developer API
COROS now has three distinct connection paths. They should not be treated as interchangeable.
| Path | Intended user | Setup | Typical capabilities |
|---|---|---|---|
| Consumer MCP | An athlete using a compatible AI assistant | No code; connect and authorize your own account | Read authorized personal training and health data in chat |
| Self-service developer access | A developer building a personal app or integration | OAuth 2.0; COROS says no application approval is required | Read 22 documented data fields and write training plans |
| Commercial partner API | A platform serving multiple COROS users | Application and review | Multi-user OAuth, webhooks, two-way integrations and enterprise use |
The approval-free language applies to the self-service developer path, not to the commercial platform route. The consumer MCP connection is also limited to your own athlete account; COROS says it does not expose coached athletes through this connection.
How to connect COROS to ChatGPT
COROS currently documents this route in ChatGPT:
- Open ChatGPT Settings.
- Select Plugins.
- Search for COROS and choose the plus icon to connect it.
- Continue to COROS, sign in and authorize the requested access.
Interface labels and account availability can change, so use the current COROS connection guide if your menus differ.
How to connect COROS to Claude
For Claude Desktop, COROS documents this flow:
- Open Settings and select Connectors.
- Open the Discover tab.
- Search for COROS and select Connect.
- Sign in to COROS when prompted.
- Review the permissions and save the connection.
COROS currently recommends Claude Pro for this use case. Free AI accounts can have connector or usage restrictions, while COROS itself does not describe a separate MCP subscription.
What can you ask it?
The connection is best suited to questions that benefit from combining several existing measurements. Examples include:
- comparing recent training load with sleep and recovery trends;
- summarizing pacing or heart-rate patterns across a block of runs;
- reviewing lap consistency from a specific workout;
- identifying questions to discuss with a coach;
- drafting a possible week structure using your current calendar as context.
Treat the output as analysis, not authority. Fitness estimates are modelled signals, and an AI answer can still misunderstand context, infer too much from incomplete data or produce an unsuitable suggestion. Pain, illness and medical decisions belong with a qualified professional.
Can AI create a COROS training plan?
It can draft a plan inside the conversation, but the current consumer MCP connection cannot write that plan into COROS. You would need to review and enter it yourself.
COROS separately documents training-plan writes for its self-service developer access. That capability does not mean ChatGPT or Claude automatically receives write permission through the consumer connector. Likewise, the commercial partner API’s two-way integration and webhook capabilities are a separate product path.
Permissions and privacy
The connection requires an explicit COROS login and authorization. Review the requested tool permissions rather than treating the connector as blanket access. Where the host supports it, individual COROS tools can be allowed, set to approval-required or blocked. That lets you withhold categories you do not want exposed, such as sleep, GPS routes or menstrual-cycle data. The exact controls vary by AI platform. Disconnect the integration when you no longer want the assistant to query the account.
Once data is returned to an AI service, that service’s retention and privacy terms also matter. COROS controls its connector and account authorization; ChatGPT, Claude or another compatible host controls how data is handled on its side. Avoid asking the assistant to process health or location data you would not be comfortable sharing with that provider.
COROS MCP is not complete secret or privacy protection. Authorization limits access, but it does not eliminate the consequences of sharing sensitive routes, health metrics or personal notes with another platform.
Current limitations
- The consumer MCP path is read-only.
- It works with your own COROS athlete data, not coached-athlete accounts.
- COROS recommends ChatGPT Plus or Claude Pro, while free accounts can have connector and usage restrictions.
- Setup and per-tool permission controls differ between supported AI platforms.
- Gemini’s standard chat does not currently support this custom connector, although COROS lists a Gemini CLI route.
- Detailed FIT/GPS access is limited to 50 file requests per day.
- An assistant can interpret data incorrectly and is not a replacement for a coach or clinician.
Who should use COROS MCP?
It makes most sense for a COROS owner who already uses ChatGPT or Claude and wants flexible questions answered from their own training history without building an integration. It is less compelling if you only want the standard dashboards in COROS Training Hub, do not want training data processed by another AI provider, or need a coach-facing multi-athlete workflow.
For the broader platform tradeoffs, see our COROS Training Hub vs Garmin Connect vs TrainingPeaks comparison and COROS vs Garmin ecosystem guide.