add_card
ChatGPTAdd a card to one of the user's own collections; the collection must belong to the caller. Provide a PSA certNumber to auto-import a graded card, OR provide card fields manually — subject is required either way. Requires a Premium or Pro subscription and the collection:write scope; get_profile reports whether writes are available (canWrite). Creates one card record and updates the collection's card count and totals, and quantity counts fully against the plan's card limit. Without a positive currentValue this may queue a background pricing job, and one research credit is charged when that job completes; if the month's research allowance is already committed the card is added un-priced and nothing is charged. A certNumber add calls PSA to look up the slab and copies its images into the collection. The subject and notes text is sent to a third-party content-moderation service, and flagged text is rejected without creating a card. If the collection has forced default traits, they override the category, set, subject and other fields you send. If the collection is already public, the new card and its image become world-readable immediately with no publish step — on the collection's public page and, unless the collection opted out, in the public gallery browse and search; notes, purchase, location and sale fields are withheld from non-owners. Returns the created card record plus any achievement themes this add unlocked; an auto-priced value lands after the response, so re-read with get_collection_card.
delete_card
ChatGPTPermanently delete ONE card from the user's collection. Called without confirm it previews the card that would be removed — its full private record, including notes, location and purchase/sale fields, plus up to 30 price-history rows; called with confirm=true it deletes. The preview is advisory, not enforced: confirm=true on a first call deletes immediately with no preview. Irreversible. The delete also removes the card's stored images (best effort — the row is deleted even if image cleanup fails), cascade-deletes its price history, nulls the card link on any eBay listing pointing at it, subtracts the card from its collection's card count and value/cost totals (a sold card already counts as zero, so nothing is subtracted for it), and logs a card_removed activity entry; a card in a public collection leaves that collection's public page at once. It spends no research credit and queues no pricing job. Both the preview and the delete are owner-scoped. Requires a Premium or Pro subscription and the collection:write scope — get_profile reports whether writes are available (canWrite).
fetch_card_image
ChatGPTFetch one side of a collection card's image and return it to the assistant as an image content block (not JSON), so it can be viewed, analyzed, or used to prepare a listing, a social post, or a grading-style review. What comes back is a downscaled ~800px re-encode, not the full-resolution original — do not treat it as evidence for fine condition or authenticity judgements. Works for any card in a collection you own and for any card in another user's public HobbyMonitor collection; private collections you do not own are refused, as are collections hidden by a moderator takedown. Call it on demand: view_card returns metadata only and never puts image bytes in front of the model. Read-only — nothing is written, deleted, or made publicly visible, and no research credit or scan quota is consumed. Requires a paid HobbyMonitor plan (Premium or Pro). Every failure returns a text block with isError set, never an image, and the message says which case it is: access denied, no stored image, rate-limited, unsupported format, too large, or an upstream error. Only the "no image is stored" message means the card actually has no image on that side.
get_card_image
ChatGPTApp-only (_meta.ui.visibility ['app'], mcp-agent.ts:343): hidden from the model and called only by the view_card UI through app.callServerTool, which delivers the result to that iframe rather than into the model's tool output. Read-only with respect to user data: one GET per call, no writes, no deletes, no cascades, and no research credit (credits are debited only on add_card's pricing-job completion path). Serves the image when the caller owns the collection OR the collection is public, so it is not limited to the caller's own cards; a moderator-hidden (galleryHidden) collection returns 'no-image' to everyone but the owner. Requires a paid tier (premium, pro or admin); free plans get isError with text 'premium-required'. width (64-1600, default 400; the UI always sends 400) selects a pre-generated webp thumbnail - width<=200 takes the 200px variant, 201-400 the 400px variant - and any larger width, or a missing thumbnail object, falls back to the stored original, edge-resized when zone Transformations are available. Success returns a single image content block (base64 data + mimeType image/jpeg|png|webp|gif, at most 700000 bytes) and no structuredContent. Failures return isError with one fixed sentinel the viewer branches on: premium-required, auth-error, rate-limited, no-image, unsupported, too-large, error; upstream error text is never forwarded. Side effects worth knowing: each call spends the caller's 200-per-minute gateway budget for /v3/collections/images (exhaustion surfaces as 'rate-limited'), a denied private-collection request writes a server-side audit warn line (cardId, collectionId, requesting uid, owner uid, IP), and a public collection's bytes are returned with 'Cache-Control: public, max-age=300' plus a per-card Cache-Tag, so they can persist in shared caches for about five minutes after a takedown (an owner's private image uses 'private, max-age=3600').
get_collection_card
ChatGPTGet one card from the user's own collection by its collection-card id (ids come from get_collection_cards). Returns {card, priceHistory}: the full stored record — every column, including purchase price/date/source, free-text notes, storage location, condition/grading and sale fields — plus up to 30 price-history rows, newest first. All of that private detail enters the conversation. Pricing metadata (pricingNotes, priceConfidence, priceCompTier, priceSaleCount, priceRangeMin/Max, isRefreshing) appears only when a cached estimate matches the card's identity hash and condition group; that estimate is shared by all cards with the same identity, is returned even when past its expiry, and is not necessarily where currentValue came from. currentValueSource reports the newest price-history source — 'manual' also when the card has no history at all — and priceDismissed is true only for a dismissed automatic value. The lookup is scoped to the caller's own uid, so an unknown id or another user's card comes back as a "Not found" error rather than data. Read-only: it reads cached pricing only, never queues pricing work, never spends a research credit and never writes to the card. Available on all plans.
get_collection_cards
ChatGPTList cards across ALL of the user's own collections (never another user's cards, never the public gallery). There is no per-collection filter and rows carry no collectionId, so this tool cannot answer "what is in collection X" - use get_collections for the collections themselves. Rows are a SUMMARY: id, year, brand/product/set, subject, parallel/variation, card #, print run/serial, team, autograph/relic/rookie flags, grading company + grade, currentValue (a decimal string), quantity, isSold. Empty fields are omitted, so a missing key means nothing stored; false and 0 are still returned. currentValue is just the stored number - a price the user typed, an automatic price, one the user dismissed and one being refreshed all look the same here. Call get_collection_card with a row's id for the full record (purchase price/date, notes, location, value source and dismissed/refreshing state, and up to 30 recent price points) - also free on every plan. search, category, year, hideSold, graded, sort and order all run server-side; filter here rather than paging and narrowing yourself. Paged: max 100 rows per call, pass offset to continue; the result reports total and hasMore, so never assume the first page is everything. Read-only: it writes nothing to the collection, queues no pricing work and spends no research credit; available on all plans.
get_collections
ChatGPTList the collections owned by the authenticated user, most recently updated first. Takes no arguments and is not paged: it returns every collection in one call, or an empty array if the user has none. Each entry is the full stored row — id (the handle add_card takes as collectionId), firebaseUid (always the caller's own), name, description, cardCount, totalValue and totalCost as decimal strings rather than numbers, isDefault, theme, treatment, linkedSetId/linkedCardSet, defaultTraits with forceDefaultTraits (traits auto-applied to cards added there), autoPricingEnabled (stored but currently inert — it does not predict whether adding a card spends a research credit), and the visibility flags isPublic, hidePricingOnShare, galleryOptOut and galleryHidden. isPublic is the field worth reading before acting: a collection with isPublic true is readable by anyone at its public link, and every card in it is listed there with no per-card publish step, so a card added or edited later shows up immediately — unless galleryHidden is true, which hides the page from non-owners. Read-only: it performs no writes and consumes no research credits. Available on all plans; no subscription required.
get_profile
ChatGPTReturn the authenticated user's own HobbyMonitor account: username, email, emailVerified, subscription tier (free, premium or pro; admin accounts report admin), the OAuth scopes this connector holds, and canWrite. canWrite is true only when the grant includes collection:write AND the tier is premium, pro or admin — the same two conditions add_card, update_card and delete_card enforce, so check it before calling them. Tier is re-resolved on every call rather than fixed at connect time, but it comes from a subscription cache that can lag a few minutes, so a just-completed upgrade may not show immediately; scopes are fixed when the user authorizes the connector and change only if they disconnect and reconnect. Takes no arguments and returns only the caller's own account, never another user's. Read-only: changes nothing and consumes no research credits. Available on all plans.
get_prospect
ChatGPTFull detail for ONE prospect by id (ids come from search_prospects): scouting grades and profile, recent news, injury/roster status, HobbyMonitor scores and 1st Bowman auto pricing, plus stat lines broken out BY SEASON AND LEVEL (all seasons on file, newest first; includeStatHistory=false skips them), so a mid-season promotion is visible. Mind the scales: fairValue is a 0-100 Value Score, NOT dollars; hmHobbyRating is 0.5-5.0 stars; communityHobbyRatingAvg averages 1-5 star collector votes; momentumScore is signed (about -100 to +100, 0 = flat); riskScore is a Low/Medium/High band, not a number. Only bowmanFirstAutoRawPrice and bowmanFirstAutoPsa10Price are dollar amounts. Reads the shared prospect catalog, not user-owned data - only isWatched reflects the caller. Any id resolves, including archived prospects that search_prospects never returns. For graduated or inactive prospects the HM rating, fair value, risk, community rating, scouting profile and news come back empty. Null or empty fields are omitted, so a missing key is not a zero. Does not carry the search row's season stat totals, 30-day price change or weekly sale counts. Read-only: two GETs, no writes, no research credits, nothing published. Requires a Premium or Pro subscription.
search_prospects
ChatGPTSearch and RANK the shared HobbyMonitor prospect catalog — global data, not the caller's collection or watchlist. FILTER AND SORT ON THE SERVER, not by paging the whole pool: "above $125 with the most home runs" is ONE call (minPrice=125, playerType=bat, sort=homeRuns). minPrice/maxPrice and sort=price use the raw 1st Bowman base auto price, in dollars. Every sort defaults to ascending, so pass order=desc for the highest price, rating or momentum; the season-stat sorts already default to best-first on their own (homeRuns/stolenBases/avg/ops/strikeOuts descending, era/whip ascending). sort=strikeOuts ranks PITCHER strikeouts, and sort=avg/ops/era/whip rank only players past a sample floor (100 AB, 30 IP) — everyone under it sorts last. Passing search without an explicit sort orders by name relevance and ignores order. Stats are season totals summed across every level played, with those levels per row; perfPlus and minPerfPlus ignore season and always use the current year. Defaults to active; search also returns graduated and inactive players, whose rows omit hmHobbyRating, fairValue, riskScore and the community rating. Note that fairValue is a 0-100 HobbyMonitor Value Score, NOT a dollar figure, and momentumScore is signed (negative means cooling). Rows are a SUMMARY — call get_prospect with a row id for scouting grades, news and the per-level breakdown. Read-only: no writes, no research credits, nothing made public. Requires a Premium or Pro subscription.
update_card
ChatGPTUpdate fields on one card in the user's own collection; the route only matches cards owned by the caller. Only the fields you send change: an empty string clears a text field, and over-long text is silently truncated. Subject and notes are sent to an automated content-moderation service before the write, and flagged text rejects the update. A positive currentValue is recorded as a price the user typed and writes a price-history row; currentValue 0 clears the card's value. Changing identity fields (subject, year, productName, cardNumber, parallelName, printRun, hasAutograph) recomputes the card's pricing identity; when that identity actually changes, the card's pending pricing jobs are cancelled and — unless the card carries a price the user typed — a re-price is queued that can later replace its automatic value. No research credit is charged. Changes to value, purchase price or quantity adjust the collection's card count, total value and total cost, and every successful call writes an activity-log entry. If the collection is already public, the edit appears on its public page immediately (notes, location, treatment and purchase/sale fields stay owner-only). Returns the full updated card, including owner-only fields such as notes, location and purchase/sale data, plus an advisory parallelSuggestion that is never applied. Requires a Premium or Pro subscription and the collection:write scope — get_profile reports canWrite.
view_card
ChatGPTShow one card in the card viewer and return that card's full record to you as JSON: identity (year, brand, set, subject, card number, parallel, print run), grading, quantity, imageUrl, plus the owner's private fields — purchase price/date/source, purchase notes, location, notes, for-sale flag, sale price/fees — and cached pricing metadata (currentValue, currentValueSource, priceDismissed, priceConfidence, priceCompTier, priceSaleCount, priceRangeMin/Max, isRefreshing). The lookup is filtered to the caller's own uid, so a cardId in anyone else's collection — including a public one — returns not found. Price history is not included; use get_collection_card for the 30 most recent price rows. Read-only: it modifies no card, collection or value, queues no pricing job, and spends no research credit. Available on all plans; the card images shown inside the viewer are fetched separately by the viewer and require Premium or Pro. Clients that cannot render the viewer still receive the JSON.
add_card
Claudedelete_card
Claudefetch_card_image
Claudeget_collection_card
Claudeget_collection_cards
Claudeget_collections
Claudeget_profile
Claudeget_prospect
Claudesearch_prospects
Claudeupdate_card
Claudeview_card
Claude