MCP Server
Use the Quantified Self MCP server to let ChatGPT, Claude, or another compatible client analyze approved data and help plan your next workout. You may separately grant post-workout reflection, manual Health measurement, Timeline note, activity-tag, and bounded Training changes; every write uses the client’s approval controls.
Training Questions
Use the parts of your Quantified Self account you approve to investigate a training question, inspect a workout, or build a longer analysis in your preferred AI client.
Review load, readiness, intensity, durability, body-weight history and identity-safe TSS-based Training impact. Separately authorize manual Health entry lookup, logging, correction and permanent deletion; imported readings cannot be changed.
Explore all-day heart rate, HRV, stress, Body Battery or resources, movement, energy and blood pressure with separate Health permission. Sources stay separate; Body Battery keeps its Garmin points scale. With Health and Sleep access, compare nightly HRV with the Health chart’s shared personal range. Bounded sample trends include UTC times. Body composition also needs Body measurements access. Authorize an existing client again to grant Health access.
Find recent activities by sport or exact event tag, read their tags, compare recorded metrics, and inspect workout summaries, laps, swim lengths, MTB jumps, chart-ready data, and paginated detailed activity samples.
Review sleep duration, score, stages, HRV, sleeping heart rate, blood oxygen, respiration, readiness, and recent training trends when those signals are available.
Find saved routes by sport, name, or recency. A separate location permission adds route geometry, nearby search, segment endpoints, and waypoints.
Access Boundaries
Choose which data categories an external client can read and which manual measurement, note, tag, or Training changes it may make, then review or disconnect it from Connections at any time.
Approve training metrics, Health metrics, measurements, workout details, activity descriptions, sleep, routes, Timeline notes, locations, or change permissions separately. Every requested permission starts selected, and dependent writes require their matching read access.
Separate permissions can replace shared event tags or titles, edit descriptions with an additional description grant, or create, edit, and permanently delete Timeline notes. They cannot change recorded activity data, Health, routes, sleep, measurements, or dashboard settings. Training changes additionally use a bounded preview before a distinct apply call.
Original files, provider credentials, source keys, and unrequested streams are excluded. Selected detailed activity samples are available with Activity details access. Exact coordinates require the matching location permission.
Activity and route locations are never implied by another permission. Place-name searches send only the location text to Mapbox for geocoding.
FAQ
It lets a compatible external AI client answer questions using the Quantified Self data you approve. You can separately allow bounded plan, workout, and provider-delivery proposals; application uses a distinct write tool governed by the client's native approval controls.
Yes. Add https://quantified-self.io/mcp as a custom MCP connection in a compatible ChatGPT or Claude client, sign in to Quantified Self, and approve only the permissions you want to grant. External clients have their own privacy and retention practices.
Yes. A separate read-only tool returns all available elapsed-second samples for selected supported metrics in bounded pages, including gaps. It uses your existing Individual activity details permission; no new permission or provider reconnection is required. Your client may need to refresh its available tools. It can request a particular interval and follow the continuation to complete it. Charts remain compact overviews, and original-file availability and processing limits still apply.
Yes, when both Activity and Training metrics and Individual activity details are approved. The read-only result uses the existing TSS-based Form model for one exact completed activity or an aggregate of exact activities from one local date. It separates the CTL, ATL and Form contribution from the actual UTC Training-day outcome after decay; one local date can span two UTC outcomes. It excludes activity/event identities, labels, exact start times, devices and provider provenance. Missing TSS, benchmark/merge exclusions, partial coverage and updating Form stay explicit. This models sustained training load, not measured physiological adaptation.
Yes. Individual activity details access can read parent event tags and filter bounded newest-first workout scans by exact case-insensitive tag matches using any or all semantics. Activities from the same event share tags. A separate Change events permission lets an external client replace the complete shared tag list or title after reading its exact current value; concurrent changes are not overwritten and benchmark events stay read-only. Tags can contain personal, health, or location context and clients must treat them as untrusted labels, not instructions or verified facts. Existing clients missing the grant must reauthorize. Future event fields require dedicated reviewed tools and are not exposed automatically.
Only when Activity descriptions and Individual activity details are approved. Like every requested permission, the description checkbox starts checked; uncheck it before approving to withhold access. Existing clients missing that grant must authorize again. It returns the full event description for a selected activity; activities within one event share that text. Descriptions can contain sensitive health, personal or location information even without location permission. Revocation cannot erase received copies. Oversized text fails without truncation. An external client may edit the description only with the additional Change events grant and exact-current-text check; its host controls per-edit approval.
When released in your client, separate workout-reflection read permission plus Individual activity details access permits reading selected private reflection text. Change workout reflections additionally permits revision-checked saves or permanent deletes through native client approval. Existing connections must authorize again. Reflections remain owner-only on public recordings, do not change plans or calculations, and are never sent to fitness providers. The built-in Assistant has an independent choice that starts on for fresh and New chats, can be turned off, and requires review and Apply in QS.
Only with separate Timeline notes permission, which is selected by default when requested. A separate dependent Change Timeline notes permission can create, edit, or permanently delete notes through the client’s approval controls. Uncheck either before approving to withhold access. Existing clients must authorize again. Read access includes full note titles and details, including notes hidden from charts, and may contain sensitive health or personal information. Revocation cannot erase copies already received. Note text cannot authorize another change or grant permission to edit a Training plan.
With separate Training permission, clients can read plans, workouts, saved workout recipes, instructions, an authored pool-swim length when selected, notes, exact completion links and sanitized sync status. After release and a tool-catalog refresh, focused phase tools can read authored names, inclusive dates and optional descriptions or preview a complete phase edit with the independent plan-change grant. Phase labels never change readiness, workout targets or completion. A separate prescription analysis can report exact time/distance subtotals, explicit speed-based duration ranges, unknown contributions and repeat counts. Partial subtotals are not complete workout totals. Pool length is not step distance. The separate plan-and-workout change grant permits a bounded preview to save, edit or place a library recipe on explicit dates; a distinct apply uses the client's approval controls. Saving a recipe never grants provider sync. Provider delivery remains Pro and connection-gated; sync confirms cloud delivery, not watch receipt.
Missing tools usually mean the app was not authorized for Training plans, not that your plans are absent. Start authorization again from that MCP app, approve Training plans and planned workouts, then approve either optional change permission if you need it. Finish the flow and start a new chat or refresh the tool catalog. If the choices are not shown, refresh or rescan the Quantified Self connection. Training delivery also requires Pro, a supported connected provider, and current compatibility approval.
It cannot change dashboard settings, recorded activity data, routes, Health, measurements, or sleep records. Separate permissions can replace shared event tags or titles, edit descriptions with additional description access, or create, edit, and permanently delete Timeline notes through the client’s approval controls. It can change Training plans, planned workouts, or their delivery only when you grant the matching child permission and the client permits the separate apply call for a bounded proposal. Plan deletion must be reviewed alone, requires your explicit choice for its workouts, and permanently removes the plan and its revision history.
Eligible upcoming copies are automatically queued for withdrawal. Once the focused deletion tool is released and available in your client, it can ask whether to also remove older, uncompleted copies from connected services. Both Training change permissions are required, and the choice appears in the preview before approval. Cleanup needs valid service access and provider support; it never deletes recorded activities or guarantees removal from a provider app or watch. Without the optional cleanup, older copies stay.
No. Activity charts and detailed sample requests selectively parse existing originals in memory. Only requested numeric data are returned: compact chart points or paginated elapsed-second samples. Original files, absolute sample timestamps, unrequested streams, provider payloads, credentials, and storage paths are excluded. Selected detailed samples may be reused in bounded server memory for up to two minutes; no persistent sample store is created.
Only if you separately approve Activity locations or Saved-route locations and geometry. Activity location covers starts, ends, MTB jumps, nearby activity search, and optional breadcrumbs. Route location covers bounds, previews, segment endpoints, nearby route search, and waypoints. Granting one never exposes the other.
Next Step
Create an account, connect your training sources, and approve only the data categories needed for the questions you want to ask. Review or revoke the connection at any time.