MCP server + agent skill
The Shopify MCP server for competitor data
Connect your RivlWatch account to Claude, ChatGPT or any MCP-enabled agent — sign in, no key to paste. Ask about rivals' live Shopify sales, read any tracked store's Meta ad creatives, and mine your swipe file — in plain language.
What is a Shopify MCP server?
MCP (Model Context Protocol) is the open standard that lets an AI assistant call real tools instead of guessing. A Shopify MCP server gives your AI live Shopify data to reason over — so answers come from numbers, not from the model's memory.
Two very different things share this name. Shopify's official dev MCP (at shopify.dev) manages your own store— admin, catalog, docs. RivlWatch's MCP server is the other direction: it gives your AI other stores' data— the orders RivlWatch observes on any tracked Shopify store, that store's live Meta ad creatives with Meta's disclosed EU reach, on-demand AI reads of any ad, and your saved-ads swipe file. If you want to manage your own store, use Shopify's. If you want your AI to answer “what is actually selling, and with which ads?” — that is this one.
Set up in 2 minutes
1 · Add the server
In Claude, ChatGPT or any MCP client: add rivlwatch.com/api/mcp as a custom connector.
2 · Sign in
Click Connect and sign in with your existing RivlWatch account — no key to create or paste. Access is revocable per connection and stops the moment a subscription lapses.
3 · Just ask
“Which of my rivals sold the most today?” · “What is allbirds running on Meta right now?” · “Sweep the Ad Library for my niche.” The AI answers from your live account.
Claude Desktop, ChatGPT, Grok, Manus — Settings → Connectors → Add custom connector
https://rivlwatch.com/api/mcpPaste the URL, click Connect, and sign in with your RivlWatch account. Claude connects from Anthropic's infrastructure rather than your machine, so there is nothing to install and no config file to edit.
Claude Code — one command, then sign in
claude mcp add --transport http rivlwatch https://rivlwatch.com/api/mcpOpens the same browser login. Cursor, Windsurf, Gemini CLI and Codex CLI take the URL the same way.
What the AI gets access to
Twenty-three tools — every one requires an active subscription or trial; there is no free tier. The 3-day trial gets full access (10 stores, 60 ads synced per store, 10 AI reads/hr) so you can test the whole surface before paying. Twenty tools are strictly read-only and unmetered on every plan. The two explicit actions are track_store and save_ad_creative; the latter requires confirmSave: true. analyze_creative is the only tool with a running AI cost — bounded by an hourly budget that scales with your plan.
| Tool | What it does | Starter$49/mo | Pro$89/mo | Agency$199/mo |
|---|---|---|---|---|
| whoami | Your account email, plan tier, subscription status | ✓ | ✓ | ✓ |
| list_my_stores | Your tracked stores: today’s revenue + orders, lifetime orders, product counts, brand, country | 10 store slots | 30 store slots | 100 store slots |
| store_creatives | One tracked store’s full Meta ads library: headline, body copy, CTA, landing page, format, days running, EU reach, media URLs | 50 ads synced /store | 100 ads synced /store | 200 ads synced /store |
| saved_ads | Your saved-ads swipe file: collections, labels, copy, targeting, archived media, cached AI creative reads | 100 saves /mo | 500 saves /mo | Unlimited saves |
| saved_tiktok_finds | Your saved TikTok Shop products (frozen snapshots) and monitored TikTok videos (plays since saved) | ✓ | ✓ | ✓ |
| my_top_ads | Reach-ranked creatives across your tracked stores, or the global pool; per-country ranking for EU markets | ✓ | ✓ | ✓ |
| ad_audience | One ad’s EU demographics from Meta’s DSA disclosure: gender split, age bands, top countries | ✓ | ✓ | ✓ |
| winning_creatives | The global winning-creatives discovery feed, filterable by window, format, country, AI-generated | ✓ | ✓ | ✓ |
| search_ads_library | Keyword search of the Meta Ads Library via our own Meta edge — ranked advertisers plus a bounded set of real, lightweight creative previews for deep agent sweeps | 120 searches /hr | 120 searches /hr | 120 searches /hr |
| get_ad_creative | Expands one selected search preview into the exact creative: full copy, lifecycle, Meta source proof and every live image/video URL. Read-only; nothing is archived | ✓ | ✓ | ✓ |
| winning_products | Products ranked by REAL observed sales across the tracked universe — velocity, sell-outs, revenue, composite score; scope to your stores or the whole pool | 24h–30d windows | 24h–30d windows | + 60d & 90d windows |
| winning_stores | Stores ranked by observed order growth over a window — who is scaling right now | 24h–30d windows | 24h–30d windows | + 60d & 90d windows |
| store_orders | One store’s observed order history over a period — the timeline behind the headline number | ✓ | ✓ | ✓ |
| product_detail | One product’s price history, stock/sell-out signals and sales trend — optionally with the live ads pointing at it | ✓ | ✓ | ✓ |
| repeat_index | Which of a store’s products look like repeat/replenishment buys rather than one-offs | ✓ | ✓ | ✓ |
| product_feed | The chronological feed of new products, restocks and price moves across tracked stores | ✓ | ✓ | ✓ |
| alerts | Alerts that recently fired on your account: rising products, new creatives and other watch triggers | ✓ | ✓ | ✓ |
| download_media | The permanent, downloadable URL for one archived creative (image or video), verified retrievable for your account | ✓ | ✓ | ✓ |
| save_ad_creative | Explicitly saves one selected search creative to your swipe file and archives its media on demand; requires confirmSave: true | uses saved-ad quota | uses saved-ad quota | uses saved-ad quota |
| store_competitors | Brands competing with one tracked store in its niche, with how many Meta ads each runs — served from cache, counted via our own Meta edge | ✓ | ✓ | ✓ |
| store_bestsellers | Per-product sales ranking for ANY Shopify store: its real best-selling sort order, review depth + rating per product, and a repeat-purchase flag | ✓ | ✓ | ✓ |
| analyze_creative | AI reads one creative: hook, angle, story arc, closing CTA, aesthetic, summary and the video transcript — fresh reads draw on an hourly AI budget; cached re-reads are always free | 20 AI reads /hr | 45 AI reads /hr | 120 AI reads /hr |
| track_store | Explicitly starts tracking a new store on your account | uses 1 of 10 slots | uses 1 of 30 slots | uses 1 of 100 slots |
✓ = included, unmetered. The 3-day free trial unlocks the full toolset; a lapsed subscription disables the connection (and every tool) with a clear subscription_inactive error until renewed.
Security model
Connections are least-privilege
You sign in through RivlWatch's own login — the assistant never sees a password or a key. Every request re-checks your live subscription, so a lapsed plan disables the connection instantly, and you can revoke a connection from your MCP client at any time.
Rate-limited by design
All key traffic on your account shares one budget — 600 requests/hour across all keys — enforced durably across our whole infrastructure. Past it, calls return HTTP 429 with Retry-After, and well-behaved agents back off automatically.
You need an account first
Connecting requires an existing RivlWatch account with an active subscription or trial — signing in does not create one. There is no free tier and no auto-provisioning: an identity we don’t recognise, or one whose plan has lapsed, gets a clear error rather than data.
A connection can never:
- ✕ Access billing, change your plan, or see payment details
- ✕ Delete stores, saved ads, collections, or other account data
- ✕ Modify an existing store, saved creative, label, or collection
- ✕ Read another account’s data — every call is scoped to the signed-in owner
- ✕ Trigger scraping, re-syncs or discovery runs that spend your plan’s credits
- ✕ Run unlimited AI analysis — creative reads draw on an hourly budget set by your plan
- ✕ Exceed your plan: entitlement is re-checked on every single request
The RivlWatch agent skill
Bundled with the server is a ready-made skill (rivlwatch-skill/SKILL.md) that teaches your agent the proven workflows: portfolio checks, reading a competitor's ads library, mining the swipe file for patterns, and the guardrails (never retry a 402/429 in a loop, treat daysRunning as the winner signal). Drop it into ~/.claude/skills/rivlwatch/for Claude Code — it's OpenClaw-compatible too.
Frequently asked
Ready to ask your AI about your rivals?
Add the connector, sign in with your account, and start asking.