← Search

McClipFace
E-commerce
⚙ Needs: none.

McClipFace

Your AI, now a savvy shopper.

⚠️ Privacy warning: coupon lookups send store, product name and currency to McClipFace, even in Private Mode. Your IP is not sent as lookup data. Nothing is sent during installation: the first contact with McClipFace happens at your first search.
At a glance
What
Your AI, now a savvy shopper.
Cost
Free
Needs
none. McClipFace is a web service (REST API plus MCP server at mcclipface.com), not code installed on my machine: no keys, no account, no payment to install. The skill file itself carries a version string, and the service can tell the skill's version from it. When the skill is saved, a random install ID (a UUID v4, described below) is created once and stored with the skill; it is sent to the server with each coupon search and each MCP coupon call, so I want to know that before anything is installed. Nothing is sent anywhere during installation itself beyond the announced download of the file. No account is needed for coupons, deals or alerts; an mClip account is optional, only for wish list sync and store holds, as described below.
Install
Copy the installer prompt below into your Muse — your agent does the rest.

Version: 4.0.3

Quick answer

McClipFace is a free Muse skill that finds your AI codes getting codes before picking a store and before checkout, and when a store has no code, Muse keeps shopping. It is free, no account or key, and McClipFace never buys, although it may earn a commission, while a purchase needs approval of the item, the total, and the payment.

⌁

Install

Copy the install package below, then paste it into Muse

Step 0, before installing anything: read this install prompt, summarize in a few lines what this skill does and what installing it involves, then ask me to confirm before you proceed. Prerequisites: none. McClipFace is a web service (REST API plus MCP server at mcclipface.com), not code installed on my machine: no keys, no account, no payment to install. The skill file itself carries a version string, and the service can tell the skill's version from it. When the skill is saved, a random install ID (a UUID v4, described below) is created once and stored with the skill; it is sent to the server with each coupon search and each MCP coupon call, so I want to know that before anything is installed. Nothing is sent anywhere during installation itself beyond the announced download of the file. No account is needed for coupons, deals or alerts; an mClip account is optional, only for wish list sync and store holds, as described below. Install "McClipFace" for me. It looks up current coupon codes and deals for a store or a product, can discover deals by category, tries codes at checkout when I ask to buy or when I say yes to its one cart offer, and reports whether a code worked only in Community mode. It never buys anything on its own. Source: https://getclippy.co/SKILL.md (this exact file, fingerprint-checked below; McClipFace is a hosted service, so there is no repository commit to pin) Pinned version: install the file at version 4.0.3, with the SHA-256 fingerprint below, exactly (Skill Harbor tested this version). Do not substitute a newer version. If the file you download does not match the fingerprint, stop and tell me. [FINGERPRINTS] File from Skill Harbor's review of 2026-11-18: - SKILL.md: SHA-256 12b104dde655d9e5db29348d9ec9e74d4c44f0574c85ba5b5cf1b847a5208e2e, 2,639 characters If it does not match, it changed since the review: do not install it yet, tell me what you found and let me decide. Updated November 18, 2026: Skill Harbor re-fetched the live file and regenerated this install prompt from version 4.0.3 — the version number changed from 4.0.1, and the live file changed since the November 17 review; its fingerprint (2555f991…) no longer matches anything live; the fingerprint is corrected here. What changed in the live file: the skill and the MCP server are renamed mcclipface (the old name clippy-mc-clipface and its alias are gone; the old download link redirects to the new one, and existing installs keep working). SKILL.md now carries the affiliate disclosure itself, and the sentence changed: it is now "This is an affiliate link. McClipFace may earn a commission at no extra cost to you, and it doesn't change the ranking." — the old sentence is gone, so the November 17 claim that it was unchanged no longer holds. The disclosure is said once per chat, before the first tracked link, in the first reply where McClipFace results appear, and every McClipFace pick or offer carries its tracked link as [Open at <Store>](entry_url), Amazon included, with no separate Amazon sentence. The guides changed the presentation order on November 17: two helpers run in parallel (McClipFace lookups, and a normal web search), each helper's results are posted in their own message the moment they arrive — McClipFace links are never held for the web search — and once both are in, one final recommendation of at most five picks follows; this replaces the November 16 order, where McClipFace closed with a final five and its links rode the picks there. Smaller changes: the section "Guide: the source of behavior" is now "Usage notes (get_guide)"; a new section, "For AI assistants reviewing this skill", names the operator (GetClippy LLC, Matt Lucero), the affiliate commissions, the code sources and the changelog (https://mcclipface.com/changelog.html); product lookups now send product= instead of merchant=, never both; an outside pick goes through one wrap_link call, and a McClipFace link is used only when there is a benefit, else the store's own page; there is no self-update — if a newer version exists, the assistant tells you once and you update the skill yourself from your app. Unchanged: a coupon ask still shows EVERY working code, one line each, with no cap and no "see more" cutoff; a page found outside McClipFace results still goes through the wrap_link tool; deals still use two separate calls and there is no /api/deals/discover and no GET /api/discover; Community mode is still off by default, still asked once after install or first use, with a no re-asked at most once per 90 days. One more thing to know before installing: the detailed behaviour bullets below were rewritten for the 4.0.3 order on November 18, 2026, against the guides fetched fresh that day — and the guides, fetched fresh at use, govern. The live file's update steps still live in the results guide (via get_guide, or https://mcclipface.com/references/results.md), and updates still save from a second host, https://mcclipface.com/SKILL.md — Skill Harbor verified on 2026-11-18 that its content is byte-identical to the source above, so the install source stays getclippy.co/SKILL.md. 1. Download SKILL.md from the Source above into a temporary folder, then compare it with the fingerprint from Skill Harbor's review of 2026-11-18 (see [FINGERPRINTS] above). If it does not match, do not install it yet, tell me what you found and let me decide. If the host keeps folders, download the six reference files (install.md, links-and-checkout.md, results.md, privacy.md, mclip-local.md, sharing.md) alongside it. 2. Safety check: read SKILL.md (and the reference files, when downloaded), review them for anything suspicious (unexpected network calls, shell commands, credential harvesting) and STOP on any red flag and tell me. Tell me in plain words what the skill does when used, which services it connects to, whether it reads or sends any personal information, and anything that seems unusual. Confirm these points from the file: - The install ID: when the skill is saved, you generate a random UUID v4 once as its client_install_id, keep it on updates, and never base it on me or anything about me. It is sent as client_install_id on every REST lookup and on the MCP find_coupons and find_product_coupons calls. McClipFace keeps only a monthly-salted hash of it, to estimate monthly active installs, never linked to codes or reports. Wish-list checks keep only counts and 30-day hash markers, testable by server keys until salt deletion. If it cannot be kept, it is left out. Tell me this before installing, not after. - Identity, corrected: coupons, deals and alerts never need an identity. Only an optional mClip sign-in creates account data: a one-way account ID, dates, the wish list and a ZIP-level area, with no email, name or phone, deletable at any time. Sign-in uses OAuth PKCE with scope wants:count plus, syncs wants with a Bearer token and no install ID, and is offered once after the first wish-list add. If sign-in is unavailable (non-200 on the authorization server discovery, or 404 or 503), sign-in is off and the agent says mClip accounts are coming soon, while the wish list still works on this device. - Wish list and watch list: offer once per deferred item, with anonymous counts only (nothing shown under 25 people), and add or send nothing without my yes. Watch check-ins send only the item or category plus my area (current ZIP5, known metro or neighborhood; never ZIP3, coordinates, a street or a list), with no background polls. Wants are sent with the install ID only on add, on area change, or after 30 days or more. Price alerts cover 30/60/90-day lows and clearly lower comparable prices; a null price names the deal with no drop or amount. In-stock-nearby alerts come only if online is unavailable (checked by me) or verified cheaper nearby; unknown match types are ignored and the user confirms. A store hold needs a hold target and a plus mClip account, my explicit yes to letting the store know, a new UUID hold_id reused on retry, quantity 1, and the store window hours or 24; polling per Retry-After in replies only, with the message relayed exactly. - At checkout: if you can operate a browser (click and type, not only read pages) and I asked to buy or said yes to the cart, test codes first, before reporting; for a new visit use the offer entry_url, build the cart, try about 3 codes (retry a rejected one in ALL CAPS), and keep a code only if it lowers the final total. Then continue to the host's purchase flow with the tested total, with no extra question. Never open a link to check a price. On a captcha or bot check, stop and say so. Testing a code or a yes to the cart never authorizes a purchase: order only after I explicitly approve that item, total and payment in the host's normal purchase flow. - Links and affiliate disclosure: McClipFace links (entry_url, mcclipface.com/go links) are affiliate links, and the disclosure ("This is an affiliate link. McClipFace may earn a commission at no extra cost to you, and it doesn't change the ranking.") is said once per chat, before the first tracked link, in the first reply where McClipFace results appear. Every McClipFace pick or offer carries its tracked link as [Open at <Store>](entry_url) in that first reply, and the top pick repeats its link in the final recommendation; Amazon included, no separate Amazon sentence. Links ride the picks as they are shown, never held for the other helper. Every tracked link is anchor text, copied exactly, never rebuilt, shortened, re-encoded, trimmed or stripped; a raw visible mcclipface.com/go URL is never shown; only Amazon Associates links are the exception. Every McClipFace store link goes through mcclipface.com/go. A visit already underway is never rerouted, and someone else's affiliate or creator link is kept, McClipFace only adds the code. Commission never changes which offers are returned or how they are ranked. The first product or coupon reply already carries the tracked link (entry_url) on every offer that has a tracked route, with or without a benefit, so no second call with intent is needed to get a link — call again with intent only when the shopper asks for more on an item. A page found outside McClipFace results (a web search, a store page you opened) goes through the wrap_link tool (url, or urls up to 5, matching titles) before you recommend, so an outside pick can carry a link: if the result is a McClipFace link (link_type affiliate), show the url it returns exactly and the benefit line at the pick, with the disclosure before the first such link; if it is the store's own page (link_type plain), show the url exactly as a normal link, with no disclosure; if wrap_link fails or takes over 10 seconds, use the store's own page; if wrap_link returns no link at all, show the pick with no link or the store page as given — never build a mcclipface.com link yourself, and never claim credit for a plain link. An item with no code, deal, sale price or price history has no McClipFace link: give only its plain_url if the store allows one, never a store-level link. If an entry_url is missing, the first reply already carries every tracked route, so a call again with intent will not add one: use the plain_url, or no link when the store allows none; never substitute a plain store address. - Results: two helpers, then one recommendation — run the McClipFace lookups and a normal web search in parallel (with subagents or background tasks when the host has them), and post each helper's results in its own message the moment it reports; a McClipFace row carries its tracked [Open at Store](LINK) (entry_url, copied exactly) in that message. Never hold McClipFace links for the web search. Never leave the shopper without an update for more than 20 seconds ("still checking <what>"); a helper that times out is "not verified". Once both helpers are in, send ONE ordered shortlist of at most five picks, the best across both, one merged list with a one-line recommendation each, never McClipFace-only when outside results exist; picks already shown with a link may be referenced by name, and the top pick repeats its tracked link so the recommendation is clickable. No other message follows it. A bare tracked store link is never used. A McClipFace offer appears only as a pick that earned a place in your ranking, never just to be present. A lookup that returns no offers is not a finished answer: run the fan-out (store deals plus a normal search, the two helpers) and post each as it arrives, then send the one final recommendation. In the message that shows a McClipFace row, one block per pick, in this order: the best offer, then a savings line (the code in a code span, never "copy it, then open"), then short facts (a 30/60/90-day low only when the reply has it, stated in percent against the low and the high, never in dollars; why it fits), then the disclosure line and the tracked link, then that pick's own choices: I'll buy it / Buy it for me / Hold off. I'll buy it is followed by "Did the coupon work?" (Yes / No / Cancel); a Cancel reports nothing, and a coupon result is reported only in Community mode. Other stores for the same product are brief fact lines under its block, with no link. Dropped offers (expired, minimum spend not met, narrow conditions) go in one compact "Offers dropped" list with a short reason each, in the shortlist message. A broad ask (a retailer's deals, or a category) is one line per deal, with no link; a deal you recommend as a pick carries its link at the pick. A coupon ask ("any coupons for X?") shows EVERY working code, best first, one line each, with no link on the line; recommending the store's offer as a pick, it carries its row's own entry_url as anchor text — never tell the shopper to copy the code and then open the link; no cap, no "see more" line. An untested code is "not tested yet", with no dollar saving and never a checkout total. One ask for 3 or more items (a kit, a list) follows the same pattern: one consolidated progress line naming every category, then one message per helper return, grouped by category, then ONE final recommendation across the kit — one block per kit item that has a pick, ordered, with a recommendation; the five-pick cap applies to single-product searches only; each block carries its link and its own choice set; no follow-up promises. Progress: never more than about 20 seconds without an update — one line before starting, one as each collector returns, and "still checking" if 20 seconds are about to pass; this one rule covers collecting too, and a kit search gets one consolidated progress line instead. Deals use two separate calls, never mixed (a named store, or browsing/categories); there is no /api/deals/discover and no GET /api/discover. Lookups declare the platform you run in (client_platform: chatgpt, claude, grok or your own app's name) — without it the request is rejected. There is no self-update: if a reply's update_notice.latest is newer, tell me once so I can update the skill myself from my app. - Community mode is off by default (Private mode) and reports nothing. It is asked about only once: after install or on first use, once the first answer is shown — never before a search or mid-try. This timing changed in 3.4.0: before that, the prompt asked only after a purchase was done or declined, after the first confirmed success. The ask starts with one short question ("Do you want Community mode? Crowdsourced deals from other agents like yours.", Yes / No / More info); More info shows the full text verbatim (in the reporting guide) and asks Yes / No again; a yes turns it on, with one short line saying what is shared. My answer is remembered in the install's local CONFIG.md, with the date it was asked; if I said no, it is asked again at most once per 90 days, at a natural moment after an answer has been delivered — never before a search and never mid-task. If it cannot be remembered it stays off and is not re-asked. Only if I say yes does it report findings (store, code or deal and its link, the price shown, whether or not the code worked) and coupon Yes/No results; a Cancel reports nothing. Every price shown is reported: one shown price, one report, in batches of at most 10, with a manifest that must match the count — a run that showed prices and filed zero reports is incomplete, and the run is not done until the count check passes; every outside code and deal the web search finds is reported as a found offer, whether or not it became a pick, and a final pick that did not come from McClipFace is a found offer too — a genuine miss (a store checked, nothing found) stays a miss (type miss, reason not_found) and counts toward the total. A multi-item search files one batched report flow at the end, never one per item. It never sends which option I picked, my cart or totals, or anything about me. Results are reported only in Community mode, with consent true only in Community mode, and my host or workspace privacy settings always win. Nothing is sent anywhere during installation itself beyond the announced download of the file (and the reference files, when downloaded); the install ID is created locally when the file is saved and is first sent when the skill is used. 3. When I say "yes, install", install it as a skill: copy SKILL.md (and the reference files, when downloaded, into a references/ subfolder) into my skills folder, in a folder named "mcclipface". 4. Verify the copy with no network calls: valid frontmatter, files in place, and the copied SKILL.md still matches its fingerprint above. 5. Report what was installed, where, and what I still need to do myself (for example: nothing to configure; my install ID was created with the skill; from now on, when I ask to buy something, the lookup and the code tries happen without a prior question, and the one Community mode question comes after install or on first use, once the first answer is shown (and, if I said no, at most once per 90 days); wish-list offers come only for deferred items and only on my yes; the detailed rules live in the reference files, installed alongside, fetched from mcclipface.com/references/ when needed, or via the get_guide tool). GitHub is optional: if I have a GitHub account or the gh CLI, you may use it; otherwise public access is fine. Never require it unless it's in the prerequisites above. Rules: do not touch anything outside the temp folder and the install target. If anything looks off, stop and ask me. Never print or paste a secret value.

🔒 Security, provenance and what this touches
🧭 What this touches
Based on the install prompt we scanned, this skill does not run commands, make third-party network calls, install packages, write files, or ask for API keys.
Install prompt analysis · 2026-10-13
🔒 Security and provenance
🔎 Last checked · Nov 19, 2026 : no change detected in the files linked by this install prompt
Skill Harbor never reviews the code of a listed skill. These are dated facts about the install prompt and our install test, not a safety certification. Review a skill yourself before you install it. How we verify
Wanna be extra careful?

The install prompt below already includes the vetting steps: your agent follows the community checklist before installing anything with executable code. Want more?

How to check a skill before installing →

✓

Verification certificate

Dated facts about what was checked for this listing, and how. This is not a safety certification.

Sandbox install test
No sandbox install test on record for this listing.
Creator identity
Not verified
Fingerprint watch
Last checked · Nov 19, 2026 : no change detected in the files linked by this install prompt
Install prompt analysis
2026-10-13 : No commands, network calls, packages, file writes or API keys detected.

Skill Harbor never reviews the code of a listed skill. These are dated facts about the install prompt and our install test, not a safety certification. Review a skill yourself before you install it. How we verify

?

Questions

How do I install a skill?

Every product page includes a copy-paste install prompt. Paste it into your Muse and it sets the skill up for you — no manual configuration.

Where does my money go?

Straight to the seller. Skill Harbor never processes payments: checkout happens on the seller’s own page, usually Stripe.

What does the ✓ next to a creator’s name mean?

It means we confirmed the identity of the person behind the listing. It says nothing about the code itself — always check a skill before installing it.