MCP App Store
by KeenEthics
Pump icon

Pump MCP Server for Claude and ChatGPT

By Riker Tech

Overview

Pump is a tempo-training workout app for iPhone and Android. Connect it to Claude to ask about your training history, analyze your progress, and build or edit the workout templates you train from. What you can do: - Read your completed workouts, with every set you logged (reps, weight, tempo, rest), and filter by exercise, workout or date. - Export your whole history as a JSON or CSV file for deeper analysis, like estimated one-rep maxes, volume trends or muscle groups you have skipped. - Build new workout templates from Pump's exercise library, with sets, reps, weights, rest and tempo, and organize them into folders. - Edit existing workouts: swap exercises, change sets and weights, add supersets, rename or move them. - Change your default exercise settings and per-exercise defaults. - Search the Pump guide for how any screen, button or setting in the app works. Changes show up in the Pump app on its next sync. Claude sees only the account you sign in with. Deleting a workout always shows a preview first and needs your confirmation; your completed history is never deleted. You need a free Pump account, created in the Pump app. There is no paid plan and no API key. Pump shares no heart rate, calories or other Apple Health data with connected apps.

How to connect

  1. Claude: Open the Pump listing in the Claude connectors directory and choose Connect. Claude listing Pump exposes 16 tools here.
  2. ChatGPT: Open the Pump listing in the ChatGPT app directory and choose Connect. ChatGPT listing Pump exposes 20 tools here.

Tools

add_exercise_media

ChatGPT
A photo, a video or a link that shows how one of their exercises is done. It appears on that exercise in the Pump app after the phone syncs. Three ways in: "url", a public image, GIF or video file (Pump keeps its own copy; an animated GIF becomes a video) or a video page such as YouTube (the app opens the page in its browser, with the page’s preview image as the thumbnail); "file", which ChatGPT fills in itself for a photo or video in the chat, whether the user attached it or your code made it; or neither, and Pump returns a one-time upload link your code sends a file in your sandbox to with an HTTP PUT, the way Claude and other apps hand Pump a file. Pass only links you found and checked: Pump refuses a YouTube video that doesn’t exist. To use the photos from a handout or screenshot of an exercise program, add once per exercise with the whole handout and "crop" around that exercise’s own photo: Pump cuts it out, whichever way the handout arrives. AI apps can add up to 250 a day for one user.

create_folder

ChatGPT
A new folder in the user’s Workouts list, to organize templates (e.g. one per program). Returns its id for a workout’s folderId in save_workout.

delete_reminder

ChatGPT
One of the user’s reminders, in one call, once the user has asked for it gone.

delete_workout

ChatGPT
One of the user’s workout templates, in two calls. First call with just "id": nothing is deleted; you get a preview of exactly what would go and a one-time confirmToken (valid 5 minutes). Show the user that preview and ask them. Only if they confirm, call again with the same "id" and the confirmToken to delete it. Their completed history is kept; the template disappears from their phone on its next sync.

export_history

ChatGPT
The user’s entire training record as one file you can download and analyze with code: every completed session with every set, plus their saved workout templates. Returns a link valid for 15 minutes (no auth header needed) and the same file as an MCP resource. "json" is { exportedAt, timeZone, units, sessions: [same shape as search_history], workouts: [same shape as get_workout] }. "csv" is one row per logged set with columns session_id, completed_at, session_name, source, session_duration_seconds, exercise_id, library_exercise_id, exercise_name, set_number, set_type, reps, weight, weight_unit, duration_seconds, tempo (eccentric-pauseBottom-concentric-pauseTop seconds), rep_goal_min, rep_goal_max. Units: timeZone: the user’s own IANA time zone; every timestamp is written on its clock; completedAt: ISO 8601 in the user’s time zone, offset included; startedAt: ISO 8601 in the user’s time zone, offset included; durationSeconds: moving time of the whole session, in seconds; sets.weight: the load lifted, in that set’s weightUnit ("lb" or "kg"); absent = bodyweight / not logged; sets.reps: repetitions completed; for sides tracked separately, reps of that side; sets.durationSeconds: seconds the set’s clock ran (timed sets); sets.tempo: seconds per rep phase: eccentric (lowering), pauseBottom, concentric (lifting), pauseTop; sets.setNumber: 1-based; when the exercise tracks left/right, odd numbers are the left side, even the right; sets.setType: "warmup" | "failure" | "drop"; absent = a working set; sets.repGoalMin/repGoalMax: the target rep range, recorded only when no actual reps were logged; sets.metrics: per-set cardio/metric actuals keyed by metric id, each with its unit; sets.completedAt: when the set was logged, ISO 8601 in the user’s time zone; distanceMeters: total distance of an imported or GPS activity, in meters; source: "pump" = trained in Pump; "healthkit" = imported from Apple Health; "import:<app>" = a history import

get_exercise

ChatGPT
One exercise from the user’s library in full detail: their saved default settings (sets, reps, per-set weights, tempo, rest, countdown, trait flags) and its how-to instructions.

get_notifications

ChatGPT
Whether Pump notifications reach the user’s phones right now, overall and for each phone in "devices" ("on", "quiet" = delivered silently, with no sound or banner, "off" with a reason: "denied", "not_asked" or "no_push_device", or "unknown"), with a note on what to tell them. Also which kinds of notification they get: likes, comments (and mentions), followers, partners (tags, buddy sessions, friends finishing a workout they shared), friendsWorkouts (people they rang the bell for), groups, workoutRequests, tips (Pump’s own nudges when they have been away) and requests (replies to bugs and features they reported).

get_settings

ChatGPT
The user’s app settings: their default exercise settings (the values every newly added exercise starts from: sets, reps, rest, tempo, countdown, weight/distance units) and their training goal. Read them before changing settings so you change only what they asked.

get_workout

ChatGPT
One of the user’s workout templates in FULL detail: every exercise with all its settings (sets, reps, per-set weights, tempo, rest, countdown, trait flags, superset grouping, how-to instructions) plus workout and per-exercise notes. Read it before changing a workout or copying exercises from it, so nothing is lost. Units: each exercise’s weightPerSet is in that exercise’s own weightUnit; duration, rest and countdown are seconds; tempo is seconds per rep phase (eccentric = lowering, concentric = lifting). Each exercise’s "id" is what save_workout uses to keep it.

list_folders

ChatGPT
The user’s workout folders: id, name, emoji, visibility ("public" folders are on their profile) and how many workouts each holds. Workouts outside every folder sit at the top level of their Workouts list.

list_reminders

ChatGPT
The user’s push reminders: each one’s name, schedule in words, next few times on their clock, whether it is paused, and which AI app added it. The answer also says whether notifications reach their phone (see save_reminder).

save_reminder

ChatGPT
A push notification on the user’s phone at a time you choose, once or on a repeating calendar schedule; it appears in the app’s Settings → Notifications, where they can pause or delete it. Without "id" it is a new reminder; with "id" that reminder changes in place, keeping whatever you leave out. "start" is on the user’s own clock: its time of day is when every reminder fires, and its date the first day. "repeat" is an RFC 5545 RRULE for the days it comes back (FREQ=DAILY, WEEKLY, MONTHLY or YEARLY, with INTERVAL, BYDAY, BYMONTHDAY, BYMONTH, COUNT or UNTIL), e.g. FREQ=WEEKLY;BYDAY=MO,WE,FR, FREQ=WEEKLY;INTERVAL=2;BYDAY=SU, FREQ=MONTHLY;BYDAY=-1FR. A reminder has one time of day, so morning and evening are two reminders. Naming a workout makes the notification open it. Returns the reminder with its schedule in words and its next three times to confirm with the user, and whether notifications reach their phone: when they don’t, or arrive quietly, tell the user and how to turn them on. Up to 25 reminders per user.

save_workout

ChatGPT
Without "id", a new workout template in the user’s library. Look exercises up with search_exercises first and pass their ids as exerciseId, so each movement arrives with the user’s own settings for it; a name that isn’t in the library becomes a new exercise. State only what you mean to set: every field you leave out is filled from the library exercise, then the user’s default exercise settings, exactly as if they had added it in the app. Exercises run in array order; give members of a superset the same supersetGroup number. With "id", that workout changes in place: every change to one they already have (supersets, order, swaps, sets, weights), including one you just created, goes this way. Read it first with get_workout. "exercises", when passed, is the COMPLETE new list in order: to keep an exercise include its "id" plus only the fields you are changing (everything else stays as it is); to add one, include it without "id"; any exercise you leave out is removed. Swapping a movement = leave the old one out and add the new one. Weights for a kept exercise are in that exercise’s own weightUnit; for an added one, in the workout’s. To superset kept exercises, list them next to each other with the same supersetGroup; supersetGroup 0 takes one out of its superset. Omit "exercises" to change only the name, notes or folder. In the app each set starts at the weight the user last lifted for it; weights you change here take its place for their next session and show highlighted. A workout’s stored weights can trail what they last lifted, so describe a weight change against their history. Returns the saved workout in get_workout’s shape.

search_exercises

ChatGPT
The user’s exercise library, their own exercises plus Pump’s built-ins, by name (case-insensitive, every word must match); omit "query" to list the whole library. Each row: id (pass it as exerciseId when building a workout, so the movement comes with the user’s own settings for it), name, category (muscle group, or "Cardio"), exerciseType ("reps" = counted reps; "timed" = a hold or cardio that runs for duration seconds), owner ("user" = their own or their customized copy; "pump" = built-in untouched), and the defaults a new add starts from: sets, reps, duration (seconds), weightUnit.

search_guide

ChatGPT
The Pump guide, the help articles on using the app: where each screen, button, and setting is, what it does, and how features work. Pass several ways of saying what you are looking for in "queries", the user’s own words and the app’s likely terms (e.g. ["slow reps", "tempo", "rep speed"]). A query matches an article that contains every one of its words, so short queries of one to three words find the most. Returns up to 5 best matches in full: steps, tips, and for each exercise setting a card of where it is, its options, default, what it does in a workout, and when it shows, each with its web url and the queries that found it. Text in [brackets] is an exact on-screen label. "android": "partial" means only part of the feature is on Android yet, and "none" means it is iPhone only. When nothing matches, returns every article title by category, so you can search again in the guide’s words.

search_history

ChatGPT
The user’s completed workout history as raw data, newest first: every session they ever logged, each with its exercises and every set as logged. Filter by exercise name ("squat" = every session that logged squats), by workout name, and by date; page older sessions by passing the oldest completedAt you received as "before". For a whole-history analysis, read the file from export_history instead. Fields and units: timeZone: the user’s own IANA time zone; every timestamp is written on its clock; completedAt: ISO 8601 in the user’s time zone, offset included; startedAt: ISO 8601 in the user’s time zone, offset included; durationSeconds: moving time of the whole session, in seconds; sets.weight: the load lifted, in that set’s weightUnit ("lb" or "kg"); absent = bodyweight / not logged; sets.reps: repetitions completed; for sides tracked separately, reps of that side; sets.durationSeconds: seconds the set’s clock ran (timed sets); sets.tempo: seconds per rep phase: eccentric (lowering), pauseBottom, concentric (lifting), pauseTop; sets.setNumber: 1-based; when the exercise tracks left/right, odd numbers are the left side, even the right; sets.setType: "warmup" | "failure" | "drop"; absent = a working set; sets.repGoalMin/repGoalMax: the target rep range, recorded only when no actual reps were logged; sets.metrics: per-set cardio/metric actuals keyed by metric id, each with its unit; sets.completedAt: when the set was logged, ISO 8601 in the user’s time zone; distanceMeters: total distance of an imported or GPS activity, in meters; source: "pump" = trained in Pump; "healthkit" = imported from Apple Health; "import:<app>" = a history import

search_workouts

ChatGPT
The user’s workout templates, each with the id the other tools take. "query" matches workout names AND the exercise names inside them, case-insensitively; omit it to list every template, most recently edited first. Read one in full with get_workout.

update_exercise

ChatGPT
The user’s saved defaults for one library exercise (default weight, sets, reps, rest, tempo, countdown, trait flags, how-to instructions): what every future add of that movement starts from. Applied immediately, so set only what the user asked for, then tell them what changed. "weight" sets every set to that load; use "weightPerSet" for ramping loads. Valid keys: sets, reps, repRangeMax, duration, restSeconds, restBetweenSides, restAfterExercise, countdownDuration, weight, eccentricFirst, inverseProgress, isBodyweight, hideWeight, trackLeftRight, isFavorite, weightPerSet, repsPerSet, tempo, restMode, exerciseDescription. Weights are in the exercise’s own unit. Editing a built-in exercise creates the user’s own copy of it. Workouts keep their own copy of each exercise: change those with save_workout.

update_notifications

ChatGPT
Switch kinds of notification on or off, as the user asks (the kinds are listed under get_notifications). Turning one off stops its push; it still shows in the app’s Activity inbox.

update_settings

ChatGPT
The user’s app settings: their default exercise settings (what every newly added exercise starts from) and their training goal. Applied immediately, so set only what the user asked for, then tell them what changed. Pass just the changed keys; read them first with get_settings. Valid keys: restSeconds, countdownDuration, sets, reps, maxRestSeconds, duration, restBetweenSides, restAfterExercise, tempoEnabled, maxRestEnabled, restBetweenSidesEnabled, restAfterExerciseEnabled, exerciseType, restMode, weightUnit, distanceUnit, goal, tempo, customGoal. For one exercise’s defaults (e.g. a default bench weight) use update_exercise instead.

create_folder

Claude

create_workout

Claude

delete_workout

Claude

export_history

Claude

get_exercise

Claude

get_history

Claude

get_settings

Claude

get_workout

Claude

list_folders

Claude

move_workout

Claude

search_app_guide

Claude

search_exercises

Claude

search_workouts

Claude

update_exercise

Claude

update_settings

Claude

update_workout

Claude

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/pump"><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/pump)