Abilities API + AI Toolkit
ExperimentalClosing the loop: the Accelerate AI Toolkit
Analytics tells you what happened. It doesn't tell you what to do about it. That's the gap marketers live in: the data is clear, the next move is not. You still have to decide what to test, configure it, wait a week, and remember to come back and call the result. Then do it again.
Accelerate 4.1.3 closes that loop. Seventeen new abilities expand what your site can tell you. An experimental Claude Code plugin, the Accelerate AI Toolkit, turns those abilities into ten workflows you run by asking a question. Built for marketers, not developers.
What a session looks like
You open Claude Code and ask: "Which landing page needs the most work?" The toolkit pulls your real traffic data, finds the page with the worst bounce-to-conversion ratio, explains why it is underperforming, and proposes two or three testable improvements. You say "set up that A/B test" and it creates the test, applies your brand's design standards, and confirms with you before publishing. A week later, you ask "did it work?" The toolkit reports a 1.8× improvement in sign-ups, with a 96% probability the new variant genuinely outperforms. Safe to ship.
More questions you can ask:
"Which landing page needs the most work?"
"What should I do first this week?"
"Set up that A/B test."
"Did the homepage test actually work?"
Ten workflows, one conversation
The toolkit is not a wrapper around your analytics API. It is a set of purpose-built workflows that compose raw data into decisions. A router reads your question and sends it to the right workflow automatically; a specialised analyst agent handles complex, multi-step investigations and reports back with findings, interpretation, and concrete next steps.
| Workflow | What you get |
|---|---|
| Review | A weekly performance snapshot you can actually read in 60 seconds, not a wall of charts. |
| Diagnose | Pick a page and find out why it is bouncing, where traffic is mismatched, and what to try first. |
| Opportunities | Three ranked next actions for the week, each backed by your real data. |
| Optimize landing page | A deep-dive on one page that ends with two or three testable improvements, ready to hand off to an A/B test. |
| Test | Plan, create, monitor, review, and declare a winner. The full A/B lifecycle, inside one conversation. |
| Personalize | Audience creation and content personalization by referrer, geography, behaviour, or device. |
| Content plan | Upcoming posts proposed from what is actually working on your site, not a generic calendar template. |
| Realtime | What is happening right now: concurrent visitors, trending content, and spike investigation. |
| Campaigns | Attribution analysis: source breakdown, UTM performance, and first-touch vs last-touch comparison. |
| Learn | Reads your past A/B test results, learns what works on your site, and tailors future suggestions. |
Nothing ships without your say-so
Every workflow that creates or changes something, whether that is an A/B test, a personalization rule, or an audience, asks for your explicit confirmation first, with the proposed change shown in plain English. You cannot accidentally publish a test, and you cannot accidentally change the experience your visitors see. This is non-negotiable for us, and it is the reason we feel comfortable shipping the toolkit to marketers this early.
Under the hood: seventeen new abilities
Under the toolkit's workflows sits the Abilities API. It is the same interface any AI agent can call directly. This release expands the surface from a handful of read endpoints into a full toolkit grouped across six domains. Everything the workflows above do is available here for custom setups.
Experiment lifecycle
Take a running test from "live" to "decided" inside a single conversation.
accelerate/get-experiment-resultsBayesian results for any A/B test: conversion rates, 95% credible intervals, and a call on whether to ship it.accelerate/stop-experimentPause, resume, stop, or call the winner. No admin clicks, no reload.
Author analytics
Benchmarking contributors and desks by the numbers.
accelerate/get-author-performancePer-author metrics with direct comparison. See which contributors or desks actually move the numbers.accelerate/get-author-contentAn author's full catalog with per-post velocity. Spot who is consistent and who is on a streak.
Real-time
Answers for "what is happening right now?" without waiting on a dashboard refresh.
accelerate/get-trending-contentContent ranked by current velocity. Catch a spike while it is still moving.accelerate/get-concurrent-visitorsLive visitor count, refreshed at a sensible cadence.
Engagement
The numbers that tell you whether a page is actually doing its job.
accelerate/get-engagement-metricsBounce rate, scroll depth, time on page, and recirculation for any page, in one call.
Campaigns and attribution
Answer "where is this working, and why?" with numbers instead of guesswork.
accelerate/get-attribution-comparisonFirst-touch vs last-touch attribution side by side. See how much credit each channel earns under each model.accelerate/get-source-breakdownTraffic by source, medium, and campaign, with engagement and conversion attached.accelerate/get-campaign-performanceUTM campaign analysis that reports conversions, not just clicks.accelerate/get-landing-pagesEntry-page analysis that shows which pages convert first-time visitors.
Power user
Escape hatches when the curated abilities do not fit your question.
accelerate/query-eventsRaw event queries when the curated abilities do not fit the question.accelerate/get-event-schemaSchema introspection: what event fields are available before you query. The lookup that makes custom queries reliable.accelerate/aggregateCustom aggregations with any dimensions and metrics. Build your own dashboard answers from first principles.accelerate/get-taxonomy-performanceAnalytics grouped by category, tag, or any custom taxonomy. Spot your strongest content verticals.accelerate/search-contentFind posts by title, URL, or text. The glue that makes every other ability composable.
Also extended
A returning ability, now smarter.
accelerate/get-performance-summaryNow accepts date presets (1h, 4h, 12h, 24h, 7d, 30d, 90d) and picks the right time bucket automatically. One call, fewer knobs.
See the AI Integration docs for the full ability reference, permission scopes, and MCP setup.
Install it today
A one-command install from the Claude plugin directory is coming once the toolkit is approved. Until then, install from source:
git clone https://github.com/humanmade/accelerate-ai-toolkit.git
cd accelerate-ai-toolkit
claude plugin install ./Then connect your site:
/accelerate-connectThe command walks you through generating a WordPress Application Password and saving it securely. Run /accelerate-status to verify the connection, then ask your first question.
Requirements: Accelerate 4.1.3+ with the Abilities feature enabled, Claude Code or Codex CLI, and Node.js 18+.
Frequently Asked Questions
Good to know
Early access, real feedback
We are shipping this early so we can learn with you. Some workflows are a first cut; edge cases in your data may surface shapes we have not seen yet. If you try it, tell us what worked and what did not. Your feedback shapes what we ship next. Share what you find on the Accelerate feedback board.