Your credentials never leave our servers. Here's how Ooty's proxy architecture keeps your marketing accounts safe, and the trade-offs we've made.
By Finn Hartley
When you connect ChatGPT, Gemini, or Claude to your Google Analytics account through Ooty, your Google password never touches our systems. Neither does your Meta login, your Amazon credentials, or any other platform password.
We use OAuth, the same authentication flow that Google, Meta, and Amazon use for every third-party app you've ever connected. You approve access on their consent screen, they send us a token, and that token is all we ever see.
This post explains how we protect those tokens, what we store and don't store, and the trade-offs we've made. Including the ones that aren't in our favour.
Most MCP tools that connect AI assistants to marketing platforms ask you to manage your own API keys. Your Google API key, your Meta app secret, your Amazon credentials, all sitting in a config file on your laptop.
For developers, this is a reasonable trade-off. For marketing teams, it's a serious problem.
API keys in local config files are:
Some tools go further in the wrong direction. They store credentials in plaintext configuration files with no encryption at all. If someone gets access to that file, they have your marketing accounts.
Ooty takes a different approach entirely.
How Your Data Stays Protected
Your credentials never reach your machine. Data flows through Ooty without being stored.
Your Machine
Zone 1
Your AI assistant
A license key(revocable any time)
No API credentials(ever)
Ooty Servers
Zone 2
Validates your identity
Retrieves your encrypted credentials
Calls the platform API on your behalf
Data passes through, nothing stored
Marketing Platforms
Zone 3
Google, Meta, Amazon, YouTube
Connected via OAuth(minimum scopes)
Credentials encrypted at rest
Your API credentials never leave Zone 2. They are encrypted at rest, decrypted only for the instant of each request, and never sent to your machine.
Here's the short version: your marketing credentials never exist on your machine. Not in a config file, not in memory, not anywhere.
When you ask your AI assistant a question like "What were my top traffic sources last month?", here's what happens:
Product Lead at Ooty. Writes about MCP architecture, security, and developer tooling.
Most marketing teams use AI the same way: copy data from one dashboard, paste it into ChatGPT, Gemini, or Claude, ask a question, and hope the answer is useful. The AI does its best with a screenshot or a CSV export, but it is working with a snapshot, not your
MCP is moving from experimental to production fast. AI assistants connected to live APIs, databases, and third-party services are no longer a proof-of-concept. They're running in companies of all sizes, handling real data, right now. The security model for MCP
Charts have always lived outside AI conversations. You run an analysis, get a table of numbers, and then open a separate tool to visualize it. Ploti changes that. It is a free, open-source MCP server that renders 43 chart types as interactive widgets directly
The whole round-trip takes under a second. Your marketing data flows through our servers but is never stored. We're a conduit, not a warehouse.
How It Works
From question to answer in under two seconds. Your credentials never leave our servers.
You ask a question
"How did our organic traffic change this month?"
Ooty verifies your identity
Your license is checked instantly. No passwords, no extra logins.
Your data is retrieved securely
Ooty connects to the platform on your behalf. Your credentials never leave our servers.
The answer appears in your conversation
Insights delivered in plain language, right where you asked.
When you connect a platform like Google, you go through their standard OAuth consent screen. This is the same "Allow access?" prompt you've seen when connecting any app to your Google account.
Here's what matters:
What Each Product Connects To
Every product requests only the minimum permissions it needs. You choose which platforms to connect.
SEO
Search Performance
Google Search Console
Google Business Profile
Read-only search and listing data
Ads
Paid Advertising
Google Ads
Meta Ads
Read + write with preview-first safety
Social
Social Media
X
Read + write with preview-first safety
Video
Video Analytics
YouTube
TikTok
Read-only channel and video data
Analytics
Web Analytics
Google Analytics
Read-only traffic and event data
Commerce
E-Commerce
Amazon
Etsy
Read + write with preview-first safety
We store:
We don't store:
For a deeper look at the security model, see our MCP security guide.
You're never locked in and always in control.
Our getting started guide walks through the full setup in five minutes.
Ooty doesn't just read your marketing data. It can take real actions across Google Ads, Meta Ads, TikTok Ads, and Microsoft Ads. Pause campaigns. Adjust bids. Add negative keywords. Swap ad creatives. Boost posts.
How Write Protection Works
The AI proposes. You approve. Nothing executes without your say-so.
AI suggests a change
Your assistant proposes an action like "Pause Campaign X" or "Increase bid by 15%".
Staged in your Ooty dashboard
The proposed change appears as a pending edit in your dashboard. Nothing is sent to the platform.
Edit, approve, or reject in the dashboard
Modify the staged edit before approving, approve as-is, or reject entirely. The AI can't bypass this.
Only then does it execute
After your explicit approval, Ooty carries out the change on the platform. Not before.
The AI cannot skip ahead. Every write operation is staged as a proposal. Without your approval in the Ooty dashboard, the change never reaches the platform.
On top of the dashboard approval gate, we've added extra guardrails for the changes that could hurt the most:
It wouldn't be transparent to present this as all upside.
Trust surface. You're trusting Ooty with your marketing account tokens. We've designed the system to minimise that risk (encryption, no payload storage, narrow scopes, revocable tokens), but you are trusting a company. If you're not comfortable with that, a self-hosted architecture where you hold your own keys is the alternative.
Latency. Every tool call goes through our servers. Compared to a direct API call from your local machine, there's an additional network hop. In practice, 50-150ms of additional latency. Acceptable for interactive use, but worth knowing about for high-throughput automated workflows. For when to use direct APIs instead, see our MCP vs API decision framework.
Scope to our products. The proxy architecture only supports the platforms we've built integrations for. If you need to connect your AI assistant to an internal API or a platform we don't support yet, you'd need a different approach.
No offline use. This is an internet-connected product. It doesn't work offline.
The alternative is what most early MCP tools do: give each user instructions for setting up their own API keys with every upstream service and running local servers that hold those keys.
For marketing professionals, that creates too much friction. Setting up a Google Ads OAuth app, a Meta developer account, a YouTube API project, an Amazon PA-API application, all before seeing any value, is where most people stop.
Ooty's approach: sign up, choose your products, go through familiar OAuth consent screens, paste one URL and one key into your AI assistant's config. That's it. No API key management, no developer accounts, no local process management.
The trade-off is trusting us with your tokens. We've tried to make that trust as narrow and as safe as possible.
We publish how Ooty works because we think you deserve to understand the systems you depend on. If an architecture can only be trusted when users don't understand it, it's the wrong architecture.
If you have questions about how we handle credentials, want to see our data processing documentation, or need anything else, reach us at hello at ooty dot io. We'd rather explain the design than have you make decisions based on assumptions.