MCP App Store
by KeenEthics
Long Run Hybrid Coach icon

Long Run Hybrid Coach MCP Server for ChatGPT

By TING-CHE CHUNG

Overview

Long Run Hybrid Coach helps you maintain one current 28-day running-and-strength direction and an executable week. It connects to Intervals.icu, compares planned sessions with completed activities, and supports training conversations from available evidence. Missing or stale readings stay unknown. Review progress, set goals and availability, revise the week or cycle together, record strength work or unrecorded activities, and confirm or deny a probable activity match. Athlete-reported records stay distinct from provider evidence. The service processes completed activities from Intervals.icu -- date, sport, duration, distance, pace and training load -- and training information explicitly supplied in MCP tool calls. Optional records include strength sets and loads, goals, availability, training preferences and how you say you feel. The connected Intervals account's display name and email address are read on request to name the account before a calendar write, and are not stored. These data support running and strength training. Plan changes and their exact calendar effects are previewed before one explicit confirmation. New workouts are included when you request delivery. Records you ask to save or correct are stored directly; fresh evidence may automatically reconcile verified completed sessions. Calendar changes affect only product-owned workouts. The preview may include filling a missing Run threshold pace required for export. Delivery is reported only as Intervals.icu acceptance after read-back; it does not prove watch receipt. Incomplete approved effects can be retried without a second confirmation of unchanged content. You can correct or retract supported records and export product-held data in the conversation. Deleting all account data is a written request instead of a tool call: the service replies with the support page at https://paceandstaystrong.com/support.html#data-by-email, which asks you to email tingcctwai@gmail.com quoting your Intervals.icu athlete ID and, where you can provide one, a screenshot of your Intervals.icu settings page. Asking in the conversation submits nothing and deletes nothing; an operator completes the request manually, and that route does not depend on a client approving a tool call. That email thread is held in a mailbox rather than in the service and is deleted on request. Training state remains until deletion. Connection-platform records and a note of which unverified sources you have been told about remain until account-data deletion. Retraction does not erase copies already included in stored decision history. The service receives the arguments sent in MCP tool calls. The chosen AI interface processes the conversation and returned context under its own terms. Review the privacy policy before connecting or sharing records.

How to connect

  1. ChatGPT: Open the Long Run Hybrid Coach listing in the ChatGPT app directory and choose Connect. ChatGPT listing Long Run Hybrid Coach exposes 22 tools here.

Tools

applyCoachDecision

ChatGPT
Commit the exact plan and calendar effects from prepareCoachDecision using its proposal. Set confirmed true when its preview requires confirmation. Retry incomplete committed calendar effects with the same proposal without another yes; a missing uncommitted preview requires prepare again.

applyWorkoutDelivery

ChatGPT
Publish or withdraw the exact set from prepareWorkoutDelivery using its proposal and confirmed true. A confirmed settings correction is written and read back before workouts. Only product-owned events may be removed. Retry partial delivery with the same proposal; a missing preview requires prepare again.

clearDeliveryAttempt

ChatGPT
Call only for delivery.unresolved_delivery when the identical set cannot be retried and the athlete confirmed after checking Intervals. Requires confirmation; clears the pending attempt but does not reconcile Intervals or modify PlanState.

confirmActivityMatch

ChatGPT
Record the athlete’s explicit confirmation or denial of one probable session/activity pair reported by the latest CoachContext. Send confirmed=true when the activity is that session, or false when it is not. The pair must currently be in reconciliation.ambiguous; a confirmation marks only that session completed, while a denial leaves the session uncompleted and prevents the same pair being asked again.

confirmPrescribedStrength

ChatGPT
Call when the athlete says a planned strength session was completed. Send session_id and only named deviations; unmentioned sets stay prescribed. One step, does not modify PlanState, and is idempotent. It writes one report per movement for that session's day, replacing any earlier report for the same movement and day. Use recordStrengthExecution for unplanned, skipped, or swapped movements.

confirmSessionNotTrained

ChatGPT
Call only when the athlete says of a past planned session that they did not train it -- 那天沒練, 這堂我跳過了. It records their statement against that session so the cycle record stops reading it as planned, in this conversation and every later one. Never call it because no activity came back: an absence is a watch that was off just as readily as a session nobody did, and only the athlete can tell those apart. Do not ask them session by session either -- a day that passed without an outcome is an ordinary state, not a question. Sessions from earlier weeks of this cycle are reachable, which is where this is usually needed; today and later are refused, and a session already recorded as completed is refused. No PlanState change, no calendar effect, and repeating it is the same record. retractAthleteRecord with kind session_not_trained takes it back. Taking it back restores what the evidence alone says -- for a week that has already elapsed that is the session reading planned with any attached activity beside it, not completed; nothing rewrites an elapsed week's own status.

exportOwnerData

ChatGPT
Asked to delete their data or their account? Then this is the wrong tool and so is every other one: no tool here deletes anything account-wide, an archive is the opposite of what they asked for, and building one reads their whole history for a request that needed none of it. Call nothing, and answer in plain text: give them https://paceandstaystrong.com/support.html#data-by-email, say that page tells them what to email, and say nothing has been submitted and nothing deleted. In Chinese, give https://paceandstaystrong.com/zh/support.html#data-by-email instead; it is the same page and the same steps. Otherwise, call this when the athlete asks what this product holds about them, or for a copy of it. Returns their plan history, decisions and reported evidence, and never a credential, a fingerprint, or another athlete's data. Takes no input: the connection decides whose archive this is. Reads and returns; changes nothing.

getCoachState

ChatGPT
Call for a plain status check: current plan id, version, week and delivery summary -- zero writes, zero Intervals calls, and it cannot reconcile. Use startCoachSession instead whenever the answer needs fresh evidence.

importAthleteHistory

ChatGPT
Call when the athlete uploads training history -- a CSV export from Strava, Intervals.icu or Garmin Connect, an Apple Health export, or a .fit file. Send the file's own text as content (base64 for .fit) rather than transcribing its rows: what this stores then came from the file. Use format records only for a source that cannot be sent as text, such as a screenshot or a PDF you read yourself. Sessions already on record are skipped or merged silently; anything genuinely ambiguous comes back in needs_confirmation, and answering is re-sending the identical payload with resolutions. Imported sessions are the athlete's own evidence, never a provider actual: they complete no planned session and are never delivered.

inspectIntervalsPermissions

ChatGPT
Call when debugging a connection, and whenever the athlete asks which Intervals account is connected. Reads Settings, the calendar and the account profile once each: the first two are classified as what the provider allowed now, and connected_account names the account itself -- lead with its email. The recorded scope list is only what the token said when it was issued. Never returns provider settings, calendar contents, or credentials, and writes nothing on either side.

prepareCoachDecision

ChatGPT
Call with one small change_request whenever the plan should move -- a week adjustment, cycle reassessment, or first plan. Returns the exact before/after values to show the athlete before asking for one confirmation, and writes nothing. The preview includes replacement/removal of affected future workouts this product already delivered. publish_new_workouts also includes new ones. When preview.calendar_delivery is present it carries target_account: name that account, email first, in the same confirmation. After startCoachSession returned no_plan_state, send only change_request, with every session carrying operation "add".

prepareWorkoutDelivery

ChatGPT
Call to build the exact preview of the selected sessions before asking the athlete for one delivery confirmation; writes nothing. target_account names the Intervals account the confirmed set would be written to -- state it in the confirmation, leading with its email. If a pace workout needs a missing Intervals Run threshold pace, settings_changes shows the narrow correction covered by that same confirmation. Set withdraw: true to preview removing superseded delivered workouts instead, when a confirmed change left one on the calendar -- session_ids then names sessions whose execution.superseded_external_id names an Intervals event the current plan no longer describes.

readCoachEvidence

ChatGPT
Call when the answer needs evidence startCoachSession did not load -- the question turned out to be about something else, the athlete changed direction, or they asked about months and the read was about today. Name the group the answer turns on; naming every group the index lists costs more than never having chosen. Returns it from that same read, so the evidence is the same moment as the rest of the answer. Never a reason to start a new session: this is cheaper and consistent with what you already have.

recordActivitySummary

ChatGPT
Call when the athlete reports training that no watch or provider captured -- a pool session, a hike, a treadmill without the watch. The record is written immediately and echoed back for them to correct by restating; no PlanState change, and nothing is written to Intervals. It reaches the coach as athlete-reported evidence and never completes a planned session. One summary per sport per day: re-sending that sport and day replaces it, so two genuinely separate sessions of one sport go in as a single combined summary.

recordAthleteAvailability

ChatGPT
Call when the athlete states available or unavailable days. One step, and does not modify PlanState. Send only the stated week or recurring days; omitted days stay unchanged. Safe to retry: re-sending a statement already on record changes nothing. A changed recurring week replaces the previous one.

recordAthleteProfile

ChatGPT
Call when the athlete says where they are or which language they want their plan in. One step, and does not modify PlanState. Send only what they stated; the other field stays as it was. Restating a field replaces it. Stored once and used by every later call.

recordLongTermGoal

ChatGPT
Call when the athlete states something they are training for past the current cycle -- a VO2max, a race time, a lift -- and again when they correct one. One step, and does not modify PlanState; it outlives every cycle and is read by all of them. Restating a metric replaces the target on record for it and keeps the fields you do not send, so a correction is metric plus the new target alone. Never call it to record a current measurement, and never to change a goal the athlete did not change.

recordStrengthExecution

ChatGPT
Call when the athlete reports completed strength sets. One step, and does not modify PlanState. Send only stated values; never infer missing load or reps. Re-send the same movement/day to correct it, which replaces the earlier report for that movement; identical reports are idempotent.

recordSubjectiveState

ChatGPT
Call when the athlete says how they are feeling -- 我覺得很累, 腿還是很沉. Stores their sentence and the day it is about, and nothing else: it is not translated, not scored, and no rule fires on it. One note per day, so restating replaces that day's note, and no PlanState changes. It exists so a run of them is visible at all -- three weeks of 很累 was three separate conversations and never a pattern -- and what a run means is yours to read from the notes and their dates, not something stored here. Symptoms are not stored here. Only what the athlete stated, never how you read it.

recordTrainingPreference

ChatGPT
Call when the athlete states how they like to train -- 習慣週五品質跑, five strength sessions a week, chest-back-legs. One step, and does not modify PlanState. Restating a topic replaces the habit on record for it. Only what they stated: never store a pattern you read out of their history, and never rewrite one because recent training diverged from it. A preference is a starting point you may plan against with a reason, not a rule. A constraint that lasts one week is the week note on recordAthleteAvailability instead.

retractAthleteRecord

ChatGPT
Call when the athlete states a stored self-reported record should not stand -- 其實那天沒練 / that entry was wrong. This is removal, not correction: re-send the matching record tool. session_not_trained removes one statement that a past session was skipped, keyed by session_id. Restating the same record later re-creates it. retracted: true means the record no longer stands, nothing-matched included -- read removed and note to tell a deletion from a no-op.

startCoachSession

ChatGPT
Call before answering any today, this-week, plan, or reassessment question; the returned PlanState is the only durable memory across conversations and holds the current prescription, beside stored athlete evidence. Say what the turn is reading for in read, and the response carries that evidence and names the rest. It also carries coaching_guidance -- the training judgment to coach from -- so there is nothing to fetch separately.

Capabilities

Writes

Example Prompts

Click any prompt to copy it.

Similar Apps

Listed on MCP App Store

Listed on MCP App Store

Add this badge to your site or README. It links to this page.

HTML

<a href="https://mcp-app-store.keenethics-labs.com/app/long-run-hybrid-coach"><img src="https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg" alt="Listed on MCP App Store" width="220" height="44"></a>

Markdown

[![Listed on MCP App Store](https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg)](https://mcp-app-store.keenethics-labs.com/app/long-run-hybrid-coach)