Sport Health MCP

A local-first Huawei Sports Health analysis MCP Server. The user merely needs to provide a HUAWEI_HEALTH_* personal data export package once, and the Agent can read exercise performance, in-exercise physiological metrics, and pre/post-exercise health context, adding historical weather and air quality as needed.

This project is for exercise review and does not provide medical diagnosis or treatment advice.

Current Capabilities

• Identify Huawei personal data export directories and core JSON files.

• Tolerantly parse Huawei non-standard JSON numeric keys.

• Parse exercise summaries, GPS tracks, heart rate, cadence, speed, altitude, and training load.

• Aggregate heart rate, resting heart rate, HRV, stress, blood oxygen, and sleep stages for the day before, the day of, and the day after exercise.

• Call the Open-Meteo historical weather and air quality API and cache responses locally.

• Provide paginated data tools and a report evidence pack through MCP.

Related MCP server: Apple Health MCP

Installation

Python 3.11 or newer is recommended, and a virtual environment should be created in the project directory:

python -m venv .venv
.\.venv\Scripts\Activate.ps1
python -m pip install -e ".[dev]"

Huawei export packages are never bundled or committed; .gitignore ignores all HUAWEI_HEALTH_* directories by default.

Local Checks

When the project directory contains exactly one HUAWEI_HEALTH_* folder, it is detected automatically:

sport-health-inspect inspect
sport-health-inspect list

It can also be specified explicitly:

sport-health-inspect --export-dir "D:\HealthData\HUAWEI_HEALTH_xxx" list

Start the MCP Server

stdio is recommended for local Agents:

$env:SPORT_HEALTH_EXPORT_DIR="D:\HealthData\HUAWEI_HEALTH_xxx"
sport-health-mcp

The core form of the MCP client configuration is as follows; the actual field names depend on the client in use:

{
  "mcpServers": {
    "sport-health": {
      "command": "C:\\sport-health-mcp\\.venv\\Scripts\\sport-health-mcp.exe",
      "env": {
        "SPORT_HEALTH_EXPORT_DIR": "D:\\HealthData\\HUAWEI_HEALTH_xxx"
      }
    }
  }
}

Registering with an MCP Client

Add the aforementioned command and env configuration to a client that supports MCP stdio, then save and restart the client. Configuration file locations and outer field names may differ between clients; refer to the corresponding client documentation.

Using with Different Agents

• Same computer: any Agent supporting MCP stdio can start the same Server, but it must be registered once in each Agent’s configuration format.

• Another user’s computer: install this project, export your own HUAWEI_HEALTH_* data package, and replace the command and data directory in the configuration with your own absolute paths.

• Web or cloud Agents: they cannot access a local stdio process on the user’s computer. Supporting such clients requires a separately deployed, authenticated Streamable HTTP service, along with mechanisms for encrypted upload, user isolation, data deletion, and privacy compliance.

The project will not share the author’s Huawei data package. The MCP Server is a generic program, and each user’s data remains on their own machine.

MCP Tools

• inspect_huawei_export: Check the integrity of the data package.

• list_activities: List exercises and monitor.

• get_activity_summary: Query a summary for a single activity.

• get_activity_track: Query the activity GPS track with pagination.

• get_activity_samples: Query heart rate, cadence, speed, and altitude samples with pagination.

• get_health_context: Query multiple days of health context around an exercise.

• get_activity_environment: Fetch and cache historical environmental data.

• get_report_evidence_pack: Generate a deterministic evidence pack for the Agent to write a report.

• get_report_contract: Get fixed report sections, length, and writing constraints.

Agents are advised to call list_activities first, select an activity_id, then call get_report_evidence_pack as the priority, and produce the report strictly according to the returned report_contract. The fixed format only retains the core judgment, performance analysis, body response, and training suggestions, avoiding repetition of the complete metrics already shown by the device app. The paginated detail tools should be used only when you need to inspect the raw curves or track.

Testing

You can run the standard-library tests without installing a test framework:

$env:PYTHONPATH="src"
python -m unittest discover -s tests -v

The tests use only synthetic activity records generated by code; they do not depend on or contain any personal exported data.

Privacy Boundaries

• Raw health data is only read locally by default.

• Only the environment enrichment tool accesses the network, sending only the route center point, date, and hour range.

• MCP tools are read-only and do not modify Huawei export files.

• Remove coordinates, timestamps, device identifiers, and health metrics before sharing logs or issue reports externally.

• Do not compress the entire working directory directly when releasing source code; use a version-control export or publish build artifacts to avoid accidentally including an ignored personal data directory.

Open Source License

This project is licensed under the MIT License. Please read SECURITY.md before reporting a security issue, and do not upload real health data, tracks, location caches, or logs containing local paths in public issues.

Third-Party Services and Trademarks

• Management data is provided by Open-Meteo, with data under CC BY 4.0. Blatantly required when displaying derived data.

• The Open-Meteo free API is limited to non-commercial use and is subject to call quotas. For commercial use, its commercial interface should be used or deploy your own instance under the official license, see Open-Meteo Terms.

• This project is an independent community project and has no affiliation with, or endorsement or authorization by Huawei. Product names and trademarks belong to their respective owners.