Claude, connected to your accounts.
19 tools, 18 of them on by default. Claude reads the brands you grant and writes nothing you have not asked for.
19 tools, 18 of them on by default
Rendered from the same manifest the server registers, so this page cannot advertise a tool Claude will not be offered. The one held back needs a permission you turn on yourself.
list_brands
Every brand on the account with its connected platforms and last sync.
list_campaigns
Every campaign that delivered in a window, with its id, platform and spend.
get_campaign_metrics
Spend, ROAS, CPA, CTR and more, by account, campaign, adset or ad.
compare_campaigns
Campaigns, adsets or ads side by side for A/B calls and budget decisions.
detect_anomalies
CPA spikes, ROAS drops and spend surges, with severity and likely cause.
get_pacing_status
Whether spend is on pace against monthly targets, with end-of-month projections.
forecast_performance
A stored forecast with its confidence interval, or an explicit statement that none exists.
get_profit
Did the brand make money: contribution margin, break-even ROAS and the SKUs losing money.
get_memory
What Cresva has learned about a brand: patterns, past decisions and how they turned out.
list_approvals
What is waiting on a human decision, and what was recently decided.
discover_storefront
What a brand's agent-readable storefront publishes, and what it supports.
search_products
Search a storefront's catalogue the way a shopper would ask.
list_products
Page through a storefront's catalogue with the offer fields agents read.
get_product
One product in full, including price, currency and availability.
compare_products
Two to five products from one brand, side by side on the fields agents read.
get_trust_score
The trust signals a storefront publishes about itself, and its tier.
scan_store
Whether an AI agent can find, understand and buy from any storefront. No API key, works on stores you do not own.
run_ai_shopper
Run the Store run on any public online store: an AI shopper shops it up to payment and stops there, so nothing is bought, the store's ads and AI assistants are read, and you get a public page to share. No API key.
18 of 19 on by default. Off until you say so: propose_budget_change.
Four steps inside Claude
https://cresva.ai/mcp
Open Settings, then Connectors
In Claude, go to Settings and find Connectors. This is the same place your other connectors live.
Add a custom connector
Choose to add a custom connector and paste https://cresva.ai/mcp into the URL field. There is nothing else to fill in: Claude reads the rest from the URL.
Sign in and choose what Claude may read
Cresva's consent screen opens. It lists exactly what you are granting. The two read permissions are on by default and the one write permission is not, so a connection that never asks for it can never reach it.
Ask something in the next conversation
Try "which of my campaigns is pacing over budget". Claude picks the tool itself.
What leaves your machine, and what does not
A remote connector talks to us directly over HTTPS, holding a token you granted and can revoke from your settings at any time.
What it may touch
Only the brands you pick on the consent screen. Of the 19 tools, 18 are reads and are on by default.
No key to manage
You authorise the connection instead of pasting a credential, so there is nothing to store or rotate, and revoking it in your Cresva settings stops it at once.
Calls are rate limited, and the published numbers are the enforced ones: 10/minute anonymously per IP, 60/minute with a public key, 300/minute with an API key, and 10,000/minute across every caller for one brand. Over the limit you get a 429 with a Retry-After header.
The same tools, whichever assistant you use
Every door below reaches the same server and the same permissions. They differ only in where you paste the URL.
Questions
What is MCP?
The Model Context Protocol is an open standard for giving an AI client a set of tools it can call. Claude speaks it natively. You add the connector by URL, approve what it may read, and the tools are there in your next conversation.
Do I need an API key?
No. You sign in to Cresva from Claude and approve the connection, so there is no key to copy, paste or rotate. Access refreshes on its own and you can revoke it from your Cresva settings, which cuts it off immediately.
What can Claude see?
Only the brands you pick on the consent screen. It cannot reach an account you did not grant, and each person on your team connects as themselves rather than sharing one credential.
Can Claude spend my money or change my campaigns?
No. Nothing the connection is offered by default writes anything. propose_budget_change is not a read: it puts a suggested budget change into your approvals queue, and it cannot approve or run it, because approving and running both happen in Cresva. It sits behind a permission that starts off.
Does it work in anything other than Claude?
Yes. The same URL works with ChatGPT and Cursor, and with any client that supports remote MCP connectors. All three reach the same server and are offered the same tools.
Paste one URL. Claude does the rest.
No key to copy, no config file to edit, and nothing for us to provision. Claude registers itself the first time it reads the URL.