English | 中文 | Changelog | Setup Guide | Skill

COROS MCP

Introduction

COROS has introduced MCP (Model Context Protocol), a secure and standardized bridge that allows AI to safely access your data on COROS servers. It is also the first officially supported MCP from a major sports watch brand.

Simply describe what you need in natural language, and AI can use COROS MCP to retrieve the relevant activity and health data from your COROS account for activity analysis, report generation, health insights, and creating, updating and scheduling training plans and workouts:

• "How has my running volume changed over the past three months?"

• "How has my sleep been over the past month?"

• "I have a marathon in six weeks. Based on my training so far, help me create and save a training plan."

• "Add a new 30-minute easy run to tomorrow's schedule."

• "Shorten the run already on tomorrow's schedule to 20 minutes, without changing my library template."

• "Based on my activity records, help me generate a 2026 marathon data report image with a tech-inspired and athletic visual style."

You can also use it to explore more possibilities and unlock the full value of your data. COROS believes athletes should own their own data and have the freedom to use it in whatever way they choose, and MCP is a direct expression of that belief.

You can configure COROS MCP on your preferred AI platform, such as ChatGPT, OpenClaw, or Codex. Connection methods are listed below.

Setup Guide

For step-by-step instructions, see the COROS MCP Setup Guide.

Choose either of the following methods to quickly configure COROS MCP.

HTTP OAuth

Add the following URL to your MCP client configuration:

https://mcp.coros.com/mcp

Supports mainstream platforms such as ChatGPT, Codex, and Claude.

SKILL

Send the following command to your AI assistant to install it through conversation:

npm install -g coros-mcp

Supports OpenClaw, Workbuddy, Hermes, and other Claw-style AI assistants.

Tool List

Note: As the project continues to evolve, the MCP tool list may change. To view the latest available tools, manually refresh the MCP tool list in your AI platform.

Training changes: choose the right destination

• A multi-week plan: create or update a training plan. New plans cover 4–16 natural weeks and start between today and 14 days from today; overlapping in-progress plans require confirmation.
• A reusable template: save or edit a workout in the library. Editing a template does not change copies already on the schedule.
• A workout on a date: create it directly on the schedule, or schedule a workout already in the library. Creating or scheduling a workout accepts dates from today through 90 days from today, inclusive.
• Changing an existing day: read the details first. Edit a standalone scheduled copy with the scheduled-workout tools; edit a workout belonging to a plan through its editable execution plan. Do not create a second workout to simulate an edit.
• New structured workouts support running, cycling and trail running. Existing library workouts from other sports, including strength and swimming, can be scheduled, but this does not mean they can be created or edited here.
• Moving or removing standalone scheduled workouts, or quitting or deleting a plan, still requires the COROS App. Completed or otherwise non-editable workouts may be protected from changes.

Creation, updates and scheduling save changes to your COROS account; asking for advice alone does not authorize saving. Actual availability and detailed limits follow the refreshed tool definitions returned by your connected server.

FAQ

1. Which platforms does COROS MCP support?

MCP is an open standard for connecting external systems. You can think of MCP as a "USB port" for AI applications. Whether a platform can connect depends on whether it supports adding MCP services and whether it supports the authentication method required by COROS MCP.

We currently recommend using verified platforms first, such as ChatGPT, Claude, OpenClaw, Workbuddy, and Hermes.

If the platform you use does not yet support MCP, you can follow its future updates. Once the platform opens compatible MCP connection capabilities, you will be able to connect COROS according to that platform's rules.

2. Do I need to pay to use COROS MCP?

COROS MCP itself is free of charge. You can authorize and connect your COROS account on AI platforms that support MCP, then use the currently available features.

Please note that some AI platforms may apply membership, subscription, or quota limits to MCP, external tool connections, advanced models, or developer mode settings. These fees are charged by the corresponding AI platform and are not determined by COROS. Please refer to the latest rules of the platform you use.

3. Unable to complete authorization?

First, make sure your COROS account can log in and work normally in the COROS App, and that your account already contains data that can be viewed.

If authorization still fails, please check:

1. Whether the authorization page has fully opened, and whether your browser blocked pop-ups or redirects.

2. Whether you signed in with the correct COROS account.

3. Whether the old connection has been removed inside the AI platform and re-authorized.

If repeated attempts still fail, save a screenshot of the error message and contact COROS customer support for troubleshooting.

4. MCP URL is invalid?

The current COROS MCP uses domain redirection technology (https://mcp.coros.com/mcp), which automatically switches to the best server based on the actual region of your account. However, some AI platforms do not support domain redirects, which may cause errors or connection failures.

If you encounter an invalid URL error, copy and enter the corresponding standalone routing URL based on the actual region of your account.

5. How do I refresh the MCP tool list after COROS MCP is updated?

When COROS MCP adds or updates tools, refresh the MCP tool list in your AI platform so the assistant can see the latest available capabilities. The exact refresh entry may vary by platform.

If the new tools still do not appear, try starting a new conversation or reconnecting COROS MCP.

6. Seeing a session expired message when using the Skill?

This usually happens when you connected to COROS through the Skill method with an older Skill version. Earlier Skill versions used a session-based MCP flow that depended on the Mcp-Session-Id session header. COROS MCP now uses the latest stateless MCP flow, and the latest Skill has been updated to support it.

Update the local Skill package:

npm install -g coros-mcp@latest

After the update, reconnect COROS MCP or start the Skill login flow again, then retry your request.

7. AI cannot retrieve COROS data?

If COROS MCP has been connected successfully but AI cannot retrieve data, try the following steps:

1. Confirm that the corresponding data already exists in the COROS App. Device data must first sync to the COROS App and then upload to the cloud. If the data does not exist in the App, AI cannot read it.

2. Open a new conversation window, or send /new in a supported client before asking again.

3. Use more explicit instruction, such as: "Please call COROS MCP and check my activity records from the past 7 days."

8. Why are AI answers inaccurate?

AI response quality can be affected by model capability, prompt wording, accessible data range, and other factors.

If an answer is clearly inaccurate, try starting a new conversation, explicitly ask it to "call COROS MCP," and specify the time range and data type. For example: "Please call COROS MCP and analyze how my running training load changed over the past 4 weeks."

We recommend using models with capabilities equivalent to DeepSeek V4 Flash or higher. Models with weaker tool-calling capability may fail to use tools or may use incorrect parameters.

9. Can COROS MCP retrieve .fit files?

Yes. COROS MCP supports retrieving .fit files for activity records, enabling deeper data analysis.

This approach also helps avoid overwhelming the AI context window with large amounts of second-by-second data.

We provide two ways to retrieve files:

• Direct retrieval: Read the .fit file entity directly.

• Link retrieval: Retrieve the file download link.

Note: Whether an AI assistant can directly display or process file entities depends on the capabilities and permission controls of the AI platform you are using. If a file cannot be downloaded or read, we recommend contacting customer support for the corresponding AI platform.

Download limit: File retrieval and download-URL retrieval share a 50-file allowance per fixed 24-hour window for the connected platform user; later calls do not extend that window.

10. Who can see my data? Is it safe?

Only after you actively authorize access can an AI application access data within the authorized scope through COROS MCP. Your COROS data remains protected by the COROS account system and privacy policy. AI conversation content and model processing methods are governed by the policies of the AI platform you use. We recommend connecting COROS MCP only on platforms you trust.

Notice

Use of COROS MCP is subject to the COROS Terms of Service and Privacy Policy. Before authorizing a third-party AI platform, make sure you understand how that platform handles conversations, files, and tool-call data. We recommend connecting COROS MCP only on platforms you trust and revoking access when you no longer use it.