Skip to main content
Glama

OpenAthlete

让 coding agent 成为真正懂你的 AI 教练。

OpenAthlete 统一佳明(Garmin)与高驰(COROS)的数据和能力,并跨会话记住你的 训练历史、伤病、目标与偏好,让 AI 在完整的个人语境中理解和帮助你。

能帮你做什么

训练与恢复、伤病与风险、比赛目标与战术、营养与装备、数据分析与资料查询…… 都可以直接讨论。它不局限于一套固定命令:凡是你能想到、AI 能结合数据和工具 完成的运动相关需求,都可以先问它。

Related MCP server: intervals-icu-mcp

安装与开始

Kimi Code:在会话中输入:

/plugins install https://github.com/whz14/openathlete

Codex:在终端执行:

codex plugin marketplace add https://github.com/whz14/openathlete && codex plugin add openathlete@openathlete

Claude Code 或其他 MCP 宿主:把这句话发给 agent:

使用 uvx openathlete-mcp 配置 OpenAthlete,并按 https://github.com/whz14/openathlete 的 AGENTS.md 完成初始化。

独立 MCP 包已发布到 PyPI 和 MCP Registry, 可由 agent 按需查询。

首次安装后按宿主要求重载或重启(Kimi Code 使用 /reload),然后另开一个与运动相关的 会话,即可进入初始化向导。环境、MCP、平台授权和偏好设置由 AI 引导或代办, 无需手动编写 MCP 配置。

之后不用背命令。直接问“你能做到什么”“你能不能……”“如何……”,它会说明 能做什么、需要哪些数据或授权,并带着你完成。

使用边界、隐私与风险

  • 需要你自己的 agent 与模型:OpenAthlete 不提供模型订阅或使用额度。分析能力 取决于所选模型,易用性取决于 coding agent、可用工具与数据完整度;AI 可能出错, 重要判断请复核。

  • 数据保存在本机,不等于数据不出机:安装更新会访问代码与依赖下载源,平台查询 和同步会访问对应 API,云端模型会收到随对话上下文发送的训练、健康和记忆数据。 账号凭据由你按向导在本机填写,不要贴进聊天。

  • 平台接口有边界:部分能力使用非官方接口,存在风控、限流和失效风险;数据是否 完整也取决于平台开放能力和你的账号。高驰采用单点登录,同步或授权期间打开高驰 App/网页可能互踢;佳明账号需关闭两步验证(MFA)。

  • 写入与健康边界:课表、计划等写回平台前会先展示内容并经你确认。AI 教练不提供 医疗诊断;出现严重伤病迹象请就医。

反馈与致谢

欢迎通过 Issues 提问题、需求或使用 场景,也欢迎按 贡献指南 提交改进。

感谢 XiaoSiHwang/garmin-sync-coros 提供的佳明→高驰同步基础,以及 garth 提供的 佳明通信能力。

本仓库自有代码以 MIT 发布。面向 coding agent 的完整文档见 AGENTS.md,产品与架构约定见 DESIGN.md。

Available Tools

9 tools
create_workoutB

预览或确认创建单次课表;按 platform 路由 COROS 官方 MCP 或 Garmin API。

Inputs: workout (required), confirm_token. Returns: ok, data, meta, error

get_current_planB
Read-only

读取当前日期覆盖的唯一进行中计划及本周日志;v0 不读取平台日程。

Inputs: date. Returns: ok, data, meta, error

get_overviewB

一页查看近期训练、统一负荷、恢复和下一赛事;必要时按覆盖状态增量同步。

Inputs: days, end_date. Returns: ok, data, meta, error

get_recovery_statusA

读取指定日期的静息心率、睡眠 HRV 与缺失项;不生成虚假恢复评分。

Inputs: date. Returns: ok, data, meta, error

get_training_loadC

读取本地 Edwards TRIMP 与 PMC 趋势,并显示锚点和历史覆盖范围。

Inputs: days, end_date. Returns: ok, data, meta, error

query_activitiesC

查询 canonical 活动;detail=true 时附结构与细区间,不返回原始 FIT 或本地路径。

Inputs: days, limit, sport, detail, +2 more. Returns: ok, data, meta, error

read_memoryA
Read-only

读取一个白名单记忆文件,或以 list 枚举白名单档案;name/list 必须二选一。

Inputs: list, name. Returns: ok, data, meta, error

sync_nowA

立即对活动和日度指标各执行至多一个增量批次;长任务由调用方继续。

Inputs: no named parameters. Returns: ok, data, meta, error

write_memoryB

以模板校验、原子替换和可选哈希并发控制写入白名单 Markdown 记忆。

Inputs: name (required), content (required), allow_create, expected_sha256. Returns: ok, data, meta, error

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.5/5.0

    Scored across 9 tools

    Disambiguation4/5

    The tools mostly target distinct resources and actions. The only overlap is get_overview, which summarizes training load/recovery and may trigger sync, versus the specialized get_training_load, get_recovery_status, and sync_now; descriptions make the summary-vs-detail boundary clear enough.

    Naming Consistency4/5

    All names use lowercase snake_case and are predominantly verb_noun patterns such as get_*, query_*, read_*, write_*, and create_*. sync_now is the only non-noun object but remains clear and consistent.

    Tool Count5/5

    Nine tools is a good fit for a focused athlete-assistant surface. Each tool covers a distinct workflow, and there is no obvious redundancy or missing low-level utility.

    Completeness4/5

    Core read, sync, memory, and workout-creation workflows are covered. Notable gaps remain around the workout lifecycle (no update/delete and no explicit list/get workout) and plan mutation, though these may be delegated to platform MCPs.

    Maintenance

    ActivityMaintained
    ResponsivenessLimited data

    Related MCP Connectors

    • Adaptive running coach MCP server — training data, plans, and recovery for AI assistants.

      https://api.coachicarus.com/mcp
    • Remote MCP server for training, nutrition, wellness, and performance data with OAuth 2.0.

      https://coachwatts.com/mcp
    • tri-train is a training platform for triathletes that syncs activities from Strava and Garmin and exposes them to Claude through a remote MCP server. Ask Claude how your last block went, why your run pace dropped, or whether you are fresh enough for Sunday's race, and it answers from your real data: activities with laps and grade-adjusted pace, weekly load and the performance management chart (CTL/ATL/TSB), sleep, HRV and resting heart rate, and your race calendar with predictions. Claude can also act: create and adjust planned workouts with structured intervals, push them to your Garmin watch, add races and write notes. Every change is journaled and can be undone. Read and write access are separate OAuth scopes, and the server enforces them on every call. Built and hosted in Switzerland by a triathlete; data stays on our own servers (Swiss revDSG / GDPR).

      https://tri-train.com/api/mcp
    • TrainingPeaks MCP server for ChatGPT, Claude and any AI agent: list calendar workouts, get workout details, add planned workout, move workout, add workout from library, comment on workout, log metric, get fitness and fatigue and form in your own TrainingPeaks account. Cotlet works in the account through your Cloud Browser, a real Chrome where you sign in to TrainingPeaks yourself; Cotlet does not store your password and you need no TrainingPeaks API key. About the official TrainingPeaks API: TrainingPeaks has an API, but only approved commercial partners get it, and it is not available for personal use (checked 5 Nov 2026). Partners read and write workouts, metrics and events through OAuth, and there is no official MCP server. Setup and FAQ: https://cotlet.dev/integrations/trainingpeaks-api-mcp-server

    Related MCP Servers

    • A
      license
      A
      quality
      C
      maintenance
      An MCP server that transforms Garmin data into compact, verdict-first coaching reports for LLMs, computing metrics like CTL/ATL/TSB and ACWR to answer training readiness questions.
      14
      61 PyPI
      Apache 2.0
    • A
      license
      A
      quality
      B
      maintenance
      MCP server that connects AI agents to Intervals.icu, enabling access to training data such as activities, wellness metrics, calendar events, gear, power curves, and custom items. Includes a running coach AI template for automated post-session analysis.
      19
      1
      MIT
    • F
      license
      Not graded
      quality
      D
      maintenance
      MCP server for TrainingPeaks that reads workouts, fitness metrics (CTL/ATL/TSB), and health data, and creates/updates/deletes planned workouts. Enables AI-powered training plan management and automated adjustment based on performance, fatigue, and injury.
      -
    • A
      license
      Not graded
      quality
      B
      maintenance
      MCP server that mirrors your Garmin data into a personal database and exposes tools for health summaries, training load, muscle readiness, and race analysis, with optional chat-driven insights via stdio or HTTP.
      1
      MIT