MCP analytics
Connect an AI assistant to Tracked analytics with MCP
Add Tracked's remote MCP server, authorize selected workspaces in the browser, and let your AI client query scoped campaign analytics.
Tracked’s remote MCP server lets a compatible AI assistant query authorized campaign analytics without copying a long-lived API token into chat. Add the server to Codex, Claude Code, Cursor, or another compatible MCP client, complete browser authorization, choose the permitted workspace and scopes, and return to the client.
The current MCP surface is read-only analytics. It helps an assistant investigate performance; it does not give the client silent permission to edit campaigns or publish landing pages.
What MCP changes in the analytics workflow
Without an analytics connection, an AI assistant can explain concepts but cannot inspect your live Tracked numbers. With an authorized MCP connection, it can request supported summaries, trends, breakdowns, funnels, traffic-quality views, events, and journeys from the workspace you selected.
That makes questions such as these practical:
- Which campaign lost the most conversion rate this week?
- Did clicks rise while qualified conversions stayed flat?
- Which device or country explains the change?
- Where does the impression-to-click-to-conversion funnel lose the most traffic?
- Did bot or trust signals change for the affected campaign?
- What happened in the attributed journey for a specific opaque reference?
The assistant still needs to interpret the data. MCP supplies a governed analytics interface, not an automatic business verdict.
Connect Tracked to an MCP client
1. Open the public setup page
Tracked provides a public setup page with the remote endpoint and client-specific instructions. The page does not ask for a token or secret before login.
2. Add the remote server
Use the client’s normal remote MCP flow. Tracked uses Streamable HTTP, so the client connects to the published server endpoint rather than launching a local analytics process.
3. Authorize in the browser
When the client requests access, continue through the Tracked browser flow. Authorization uses OAuth with PKCE and checks your current workspace membership and role.
4. Choose the permitted access
Select the workspaces and scopes the client may use. Start with the smallest analytics access that answers the task.
5. Return to the assistant
After approval, the browser returns control to the MCP client. The assistant can now discover and call the read tools allowed by the grant.
Analytics tools available through Tracked MCP
The current server exposes ten read tools:
| Capability | What it returns |
|---|---|
| Workspaces | Workspaces selected for the current grant |
| Dimensions | Supported privacy-safe breakdown dimensions |
| Summary | Aggregate impressions, clicks, conversions, rates, and revenue |
| Timeseries | Daily or bounded hourly aggregate analytics |
| Breakdown | A bounded allowlisted dimension breakdown |
| Funnel | Aggregate impressions, clicks, and non-test conversions |
| Traffic quality | Bounded bot, trust, network, device, and geography signals |
| Events | Bounded event rows with identifiers redacted by default |
| Journeys | Visitor-wide journey summaries grouped into sessions |
| Journey | One workspace-bound timeline of sessions and events |
Events and journey tools are treated as sensitive. They require the appropriate scope, and direct identity is redacted by default. Journey references are opaque and workspace-bound.
A useful analysis pattern
Start broad, then narrow only where the data points.
- Read the workspace summary for the intended date range.
- Compare the time series to find when the change began.
- Ask for supported dimensions before choosing a breakdown.
- Inspect the largest meaningful campaign, page, platform, device, or geography change.
- Use funnel or traffic-quality data when the symptom suggests stage loss or invalid traffic.
- Read individual events or journeys only when aggregate data cannot answer the question.
This keeps the analysis focused and reduces unnecessary access to granular rows.
MCP access follows the workspace
An MCP grant does not become a permanent bypass around Tracked membership. Authorization remains tied to the selected workspace and the user’s current role.
You can review and revoke connected clients from Tracked settings. Revocation and role changes are access-control events; they are separate from deleting analytics data or changing a campaign.
MCP analytics and Tracked Site Agent are different
Tracked MCP is the portable analytics connection for compatible external clients. Its current tools are read-only.
The Tracked Site Agent is the in-product operating workflow. It can work with private page drafts and has separate approval-gated capabilities for files, offers, sites, and campaigns.
Use MCP when you want your chosen assistant to analyze authorized Tracked data. Use the Site Agent when the work needs to continue into a governed landing-page or campaign workflow inside Tracked.
Frequently asked questions
Which AI clients can connect to Tracked MCP?
Tracked provides setup guidance for Codex, Claude Code, and Cursor. Other clients can connect when they support the same remote MCP and browser authorization flow.
Do I need to copy an API token into the AI client?
No. The standard setup uses browser-based OAuth authorization rather than asking you to paste a long-lived Tracked token into chat.
Can an MCP client edit my campaigns?
Not through the current Tracked MCP analytics surface. The exposed MCP tools are read-only.
Can I revoke an MCP connection?
Yes. Connected-client access can be reviewed and revoked from Tracked settings.