Connect your AI client to live experiment data. Test what AI crawlers see. Build tests by conversation. Every change waits for your approval before it ships.
MCP server
45 hosted tools, 16-tool local server
Access levels
reporting, readOnly, all
Knowledge base
9523 documentation chunks across 1145 support articles
Approval model
Every write proposed for your review
Shipping used to be the hard part. It is not anymore. AI can generate variations, headlines, or entire pages in seconds. So the constraint has moved. The question is no longer whether you can build something. It is whether the thing was worth building, and whether you can tell what actually worked. Shipping more only helps if you learn more. Otherwise, it is simply noise.
Convert Experiences runs that discipline on a stats engine refined across 17 years. It offers Frequentist and Bayesian modes, sequential testing for planned peeking, multi-armed bandits, SRM checks, and outlier detection. AI in Convert Experiences sits on top of that engine rather than beside it.
And because we believe in releasing AI features with the right foundations, the AI features in Convert are built on version control, approvals, and audit trails. Those are what make an agent something you can supervise instead of something you have to convince yourself to trust.
The Convert MCP server connects your AI client to Convert data, with no middleware in between. It works with Claude Desktop, Cursor, and any MCP-compatible client. Your client reads live experiment data and acts on it in the same place you already write code.
The hosted connector exposes 45 tools - 40 safety-separated read, write, and delete tools plus five optional render tools, across experiences, goals, audiences, feature flags, and reporting. The local npm package is a compact 16-tool server. A built-in knowledge base of 9,523 documentation chunks across 1,145 URLs answers platform questions without leaving your editor.
Hosted, you sign in with Convert OAuth and approve which accounts and projects it can reach - read, write, and delete are separate tool surfaces. Local variant, three profiles apply: reporting (the default, read-only), readOnly (adds configuration and history), all (unlocks writes). Either way, your Convert account permissions are the final boundary.
The local npm server signs every request with HMAC-SHA256 and expires it after 60 seconds; your API keys never leave your environment. The hosted connector stores no keys at all - the Convert OAuth scopes access to the accounts and projects you explicitly approve.
Convert Experiences ships 20 curated prompts for jobs you repeat. They cover pre-launch QA drafts, health checks, portfolio audits, etc. Each one hands your AI client a process to follow, ensuring it doesn’t improvise. Guardrails block bad payloads and invalid update paths. Bounded retry never retries an unsafe write. Activating or pausing needs your explicit confirmation, and permanent deletion needs a second confirmation.
AI assistants and answer engines increasingly read a plain-text /llms.txt file to understand and cite a site. Those crawlers do not run JavaScript. Client-side testing never reaches them. Convert Experiences tests llms.txt server-side, so you can find out which version earns the citation.
The JavaScript SDK decides the variant on the server before anything is returned. One consistent visitor ID means a crawler hitting /llms.txt and a person hitting a page get the same assignment, so the experiment stays valid across both.
Teams test content, format, link density, and section order. Plain-text endpoints cannot run trackers, so Convert Experiences tags the links inside each variant with its variation key. Read that tag on arrival and you learn which version AI tools cited, how many sessions it sent, and what those sessions were worth.
Convert Intelligence is a chat layer built on the Convert MCP server. It brings work that used to need a Claude-plus-Cursor-plus-MCP setup into the Convert app itself. Analyze a report. Create a goal. Edit a draft variation. Ask what to test next. All of it in plain conversation. Convert Intelligence is in Beta as of 2026 and will gradually roll out to more users.
Convert Intelligence proposes, and you accept. Reads such as analysis and lookups run freely. Every write is proposed in the interface and saved only when you accept it.
The backend blocks Convert Intelligence from touching any live experience. It works on drafts and paused experiences only, and it inherits your own permissions without ever exceeding them.
Every AI-initiated action is logged with a timestamp and shows up in your change history. You can see what the AI did, when it did it, and what you approved.
The same instinct that makes Convert Intelligence propose rather than act also governs your team. With the Request Approval System, an experience requires a sign-off before it goes live. When someone tries to activate it, the action is held until a reviewer approves or rejects it.
Approval is optional and set per rule. Switch it on for activating an experience, pausing one, or editing a live one, and scope each rule to every project or only the ones that matter. Teams that want speed leave it off. Teams that need governance turn it on.
A reviewer is an Owner, Account Manager, or Admin, assigned per rule. No one can approve their own request, so every sign-off is a genuine second pair of eyes. A requester activates the experience only once the request has been approved.
Every request, approval, and rejection is written to your change history with the person and the timestamp. The Request Approval System is available on Pro and Enterprise plans. The same workflow is built to cover more request types across Convert over time.
Convert Experiences will keep adding AI-assisted features here until AI assistance is simply how experimentation works. Each addition follows the same order. The infrastructure that makes a change reviewable ships first. The AI that uses it ships second.
None of this is compulsory. Convert Experiences will never take that choice away from you. Do the work by hand, in the ways you have always trusted, and the platform supports it fully. Or let AI step in. The infrastructure is the same underneath, and so is the standard every result is held to.
How Convert builds
Convert Experiences uses first-party cookies only and stores experiment data on servers in Frankfurt, Germany. Compliance is built into the development lifecycle, not bolted on after launch.
Convert Experiences publishes its pricing in full on a single pricing page, with flat-rate plans and no per-module add-on charges. Every paid Convert Experiences customer can self-serve from day one.
Convert Experiences is an experimentation platform. Not a CMS, not a personalization-only product, not an analytics suite. Convert ships what experimentation needs and stops there.
Convert Experiences is operated by a Climate Neutral Certified business. Convert runs on carbon-neutral servers in Frankfurt, donates 1% of revenue to Stripe Climate and 1% to conservation initiatives, and uses 8% of every plan fee to plant trees through TIST, totaling over 94,035 trees planted as of May 2026.
Does Convert Experiences let AI run experiments on its own?
No. Convert Experiences uses AI to help you build, analyze, and report. Every write action is proposed first. It saves only when you approve it. The backend also blocks Convert Intelligence from touching live experiences. AI-assisted edits apply to drafts and paused experiences only.
What is the Convert MCP server?
The Convert MCP server links your AI client to Convert data. It works with Claude Desktop, Cursor, and any MCP-compatible client. As of 2026 it offers 45 hosted tools and a16-tool local server. It also bundles a searchable knowledge base of 9,523 indexed documentation chunks.
How do I control what an AI client can change in Convert Experiences?
Convert Experiences offers three MCP access levels for the local variant. Reporting is the default and is read-only. ReadOnly adds configuration and history. All enables writes. Convert recommends starting at reporting and moving up only when a workflow needs to write.
Is the Convert MCP server secure?
Every request to the Convert MCP server is signed with HMAC-SHA256. Each one expires after 60 seconds. Your API keys never leave your own machine. Access is least-privilege by default. Convert recommends a human review before AI changes reach a live test.
Can Convert Experiences A/B test an llms.txt file?
Yes. AI crawlers do not run JavaScript, so Convert Experiences uses its server-side JavaScript SDK to pick the /llms.txt variant for every request. One consistent visitor ID means crawlers and humans get the same assignment. Add UTM tags to the links inside each variant to attribute the sessions that follow.
What is Convert Intelligence?
Convert Intelligence is a chat layer built on the Convert MCP server. It reads reports, creates objects, edits variations, and guides your testing. All of it happens in plain conversation inside the app. Convert Intelligence is in beta as of 2026 and opens gradually through a waitlist.
Does AI assistance affect statistical validity in Convert Experiences?
No. The same engine checks every result, whether a human or an AI built the variation. Convert Experiences runs Frequentist or Bayesian stats, SRM checks, and sequential testing on all of them. Server-side llms.txt decisions keep random assignment and exposure intact.
Can I require human sign-off before an experience goes live in Convert Experiences?
Yes. The Request Approval System in Convert Experiences holds an activation until a reviewer approves it. Reviewers are Owners, Account Managers, or Admins. No one can approve their own request, so every sign-off is independent. Every decision is logged with actor and timestamp. The workflow is optional and is available on Pro and Enterprise plans.
AI in Convert Experiences
Convert Intelligence is being rolled out in Beta. Try it by starting a free trial. The Convert MCP server and server-side llms.txt testing are already live on every paid plan and in trial.
15 days, full features
Editorial DisclosureThis page is reviewed quarterly. The next scheduled review is October 2026. Written by Uwem Usa (freelance copywriter, working with Convert). Reviewed by Carmen Apostu, Editor, and Trina Moitra, Chief Marketing Officer at Convert. All product claims verified against convert.com, support.convert.com, and Convert’s public SDK source as of July 2026. Convert Experiences is our product. All comparisons to other tools come from public sources. Verified as of July 2026.
