Connect Claude, ChatGPT, or Gemini to your data

Attribution.ai connects directly to Claude, ChatGPT, and Gemini so you can ask questions like "How is Meta performing this week?" or "Which channel has the worst incremental ROAS?" and get answers grounded in your actual marketing data — no spreadsheets, no copy-paste.

This is the AI connector setup guide. You don't need to be a developer to follow it.

Before you start

You'll need:

An Attribution.ai account with at least one connected channel (Shopify, Amazon, etc.).
A paid Claude, ChatGPT, or Gemini account that supports custom connectors. Free tiers don't have this feature on most providers as of April 2026.

If you're not sure whether your AI plan supports custom connectors, search the provider's docs for "MCP" or "custom connector" — that's the underlying protocol Attribution.ai uses.

Easiest path — one-click connect from the dashboard

The fastest way to connect is from inside Attribution.ai:

Open Settings → API & AI in the dashboard, or visit /dashboard/settings/api.
Find the AI you want to connect — Claude, ChatGPT, or Gemini.
Click Connect.
A new tab opens to the AI provider; sign in if needed and approve the connection.
You're redirected back to Attribution.ai with a green confirmation. That's it.

This route uses OAuth, so there are no API keys to copy and no expiration to babysit.

Connecting from your phone

If you started this on your laptop but want to finish on your phone (or vice versa), the connector page shows a QR code next to each provider. Scan it with your phone's camera and the connection link opens directly on the device — including the AI provider's mobile app if it's installed.

Step-by-step in each AI client

If your client doesn't pick up the connection automatically, here's how to set it up by hand.

Claude

Open Claude (web or desktop).
Click your name → Settings → Connectors.
Choose Add custom connector.
Server URL: https://attribution.ai/api/mcp/rpc
Choose OAuth when asked for the auth method, sign in to Attribution.ai when prompted, and approve the data scopes.
Open a new chat and ask: "What tools do you have available from Attribution.ai?" You should see a list of analytics, attribution, and operations tools.

ChatGPT

Open ChatGPT and go to Settings → Connectors (Plus / Team / Enterprise plans).
Add a connector with server URL https://attribution.ai/api/mcp/rpc.
Complete the OAuth flow when ChatGPT redirects you to attribution.ai.
Start a new chat with the connector enabled and try a real question, like "How did Meta spend perform last week vs the prior week?".

Gemini

Open Gemini → Connectors / Extensions.
Add server URL https://attribution.ai/api/mcp/rpc.
Authenticate via OAuth when prompted.
Confirm by asking "List the Attribution.ai tools available." — Gemini should respond with a tool inventory.

If any of these screens look different from what's described, the providers update their UIs frequently — search the provider help center for "custom connector" or "MCP server".

What the AI can actually do

Once connected, the AI can answer questions about:

Performance — channel-level revenue, spend, ROAS, customer counts.
Attribution — which channels touched a conversion, decay, blended views.
MMM — channel contribution, response curves, budget shift simulations.
Incrementality — running and recent lift tests, iROAS, confidence intervals.
Surveys — recovered demand, post-purchase signals.
Operations — sync status, integration health, recent activity.

The AI sees the same data you do in the dashboard — nothing more, nothing less. If a teammate has read-only access in your workspace, their AI connector will be read-only too.

Quick troubleshooting

"Could not connect" or "Authentication failed" — make sure you're signed in to Attribution.ai in the same browser, then retry the OAuth flow.
"No tools available" — refresh the AI client. Some providers cache the tool list for a few minutes.
"The data looks empty" — your AI is connected, but your workspace has no data for the date range you asked about. Check Settings → Integrations to confirm syncs are healthy, or ask the AI: "What's my earliest and latest data point?"

If none of these work, email support@attribution.ai with the AI client name and a screenshot of the error and we'll dig in.

For developers

Most users can stop reading here. The rest of this page is for engineers building custom MCP clients or scripted integrations.

Endpoint

POST https://attribution.ai/api/mcp/rpc
Protocol: JSON-RPC 2.0 over HTTP
Auth: OAuth 2.1 + PKCE (preferred) or bearer API key

OAuth discovery

GET https://attribution.ai/.well-known/oauth-authorization-server
GET https://attribution.ai/.well-known/oauth-protected-resource

When to use API keys instead of OAuth

Automation scripts and CI jobs
Backend-to-backend pipelines
Clients that don't support OAuth + PKCE

Create one in Settings → API & AI and pass it as Authorization: Bearer atai_.... Keys are shown once at creation; rotate per-integration.

Tool families

Discoverable via tools/list at runtime, or browse /developers/mcp/tools in the app:

discovery and documentation: discover, explain, ask
analytics: performance, campaigns, trends, customers, surveys, products
planning and analysis: optimization, experiments, insights, compare
operations: manage, diagnose, influencers

Error codes

401 Unauthorized — token/key is invalid, revoked, or expired.
429 Rate limited — back off with exponential delay.
400 Bad request — confirm endpoint is /api/mcp/rpc and payload is JSON-RPC 2.0.
No data returned — verify the workspace has synced data for the requested range.

Security

Create one key per integration. Don't share keys across services.
Store keys in environment variables or a secrets manager. Never commit them.
Read-only scoped credentials cannot perform write actions; use them by default.
Revoke a key immediately on exposure.

Setting Up AI Connectors (MCP)