MCP Server

Backstroke MCP Server

Developer Preview

The predictive layer for email, callable from your AI assistant.

Ask your assistant what will win before you send. Backstroke returns ranked, evidence-backed predictions grounded in outcomes from thousands of brands.

Last updated: 2026
Developer Preview This page describes how the Backstroke MCP Server works and what it will do as access opens to design partners. Connection details and a quickstart will be published here as the preview expands.

Overview

The Backstroke MCP Server connects your AI assistant to Backstroke's prediction models. Most marketing connectors give an assistant read access to your own account data. The Backstroke MCP Server returns something your account cannot produce on its own: a prediction of what will perform, drawn from a cross-brand corpus of real email outcomes.

You bring a question or a set of options. Backstroke returns a ranked answer and the reasoning behind it, inside the assistant you already use.

How it works

The Backstroke MCP Server exposes Backstroke's prediction models as tools your assistant can call in plain language. The flow is the same across every capability:

  1. You ask. Describe what you are testing, such as subject lines, a hero image, or a creative direction, in natural language.
  2. Backstroke predicts. Your request is scored against models trained on outcomes from thousands of brands, not on your account alone.
  3. You get a ranked answer. The assistant returns the options in predicted order with the reasoning that drove the ranking, so you decide with context instead of a guess.

A stateless prediction needs nothing but the content you are evaluating. Richer, account-aware predictions become available as you connect more context.

// ask → predict → ranked answer

Capabilities

Backstroke's capabilities are organized as a ladder. The more context you provide, the more specific the prediction. The early rungs need nothing but the content you are evaluating. Later rungs draw on your enriched audience to weight predictions to the people you actually send to, and to publish the winner straight to your ESP.

Stateless predictions

Live in preview

No setup required. Evaluated only on the content you submit.

subject-line ranking

Submit candidate subject lines and receive them ranked by predicted performance, with the reasoning behind the order.

Subject-line ranking inside an AI assistant: five candidate subject lines ranked by predicted open index, with a top pick highlighted.
// stateless subject-line ranking · no connection required
creative + hero ranking

Submit creative options and receive a predicted ranking of which is most likely to perform, with a read on how each one indexes across audience segments.

Hero ranking inside an AI assistant: four hero images scored and ranked, with a per-cohort index table broken out by age band.
// stateless hero ranking · scored against the cohort model
Enrichment · the step between the rungs

Connect your ESP and Backstroke maps your audience onto its prediction model. This is a one-time setup step, not something you do in chat. Once your audience is enriched, every prediction below is weighted to the people you actually send to, rather than a general audience.

Account-aware predictions

Planned for preview

Once your audience is enriched, predictions are weighted to your real audience, and the winner can be published straight to your ESP.

audience-weighted winner + lift

Run the predictive funnel against your audience and get the winning creative with an expected lift over your current default, plus a per-cohort breakdown of where it wins.

The enriched predictive funnel: generate, quality filter, cohort score, and select winner, returning a winning block with a projected lift and a per-cohort lift breakdown.
// enriched predictive funnel · audience-weighted, lift-backed
per-cohort subject variants

Subject lines tuned to each segment of your audience, not a single line for everyone.

publish to your ESP

Send the winning content straight to your ESP, behind an explicit approval step, so nothing goes out without your sign-off.

What you can ask

Once connected, you can ask your assistant questions like these:

Subject lines

Rank these five subject lines by predicted performance and tell me why the top one wins.
I am deciding between these two subject lines for a winback. Which is the stronger bet?

Creative and hero

Here are three hero images for our spring send. Which is predicted to perform best?
Rank these creative directions for a launch email.

Brand and audience

Based on our brand's history, which of these subject lines fits us best?
Which of these concepts is the closer fit for our audience?

Prediction, not generation

Most AI tools generate. They write you a subject line. The Backstroke MCP Server predicts. It tells you which option is most likely to win, and why.

The difference is the data underneath. A generated suggestion is a model's best guess from language alone. A Backstroke prediction is grounded in real outcomes across thousands of brands, the kind of signal no single account contains. That is why a prediction comes back ranked and reasoned, not merely written.

Security & data handling

The Backstroke MCP Server follows the standards of the Backstroke platform:

  • SOC 2 Type II compliant. Backstroke maintains SOC 2 Type II controls across its platform.
  • Encrypted connections for every request.
  • Authentication required for every operation.
  • Role-aware access that respects your existing Backstroke permissions.
  • Approval before publishing. Writing content to your ESP always requires your explicit approval. Backstroke never publishes on your behalf without it.

Stateless predictions are evaluated only on the content you submit. Account-aware predictions require you to connect your account context explicitly.

Roadmap

The developer preview is the first rung. What is ahead:

  • Beyond Claude. The same predictive tools, callable from other AI assistants.
  • Deeper account awareness. Predictions tuned to more of your brand and audience signal.
  • Broader creative coverage. Predictive coverage across more of the assets in a campaign.

Availability

The Backstroke MCP Server is in developer preview and rolling out to design partners. Connection details and a quickstart will be published here as access opens.

Get early access