COMPARE // XI VS GROWTHBOOK

GrowthBook flags features.
Xi runs experiments.

GrowthBook is the leading open-source feature flag and product experimentation platform: SDKs in every language, a clean stats engine, self-hostable, dev-first. Xi captures the layer above the codebase: the contract behind a marketing experiment, channel-agnostic, no flags required.

MCP for your agents · unlimited archive · no card required
WHY_SWITCH //

Why teams switch from GrowthBook to Xi

GrowthBook is excellent at what it does: flag-based experimentation in code, statistical rigor, open-source ergonomics. The thing it does not do well is marketing experiments, because marketing experiments are not flags. They are contracts: hypothesis, metric, kill threshold, end date. Teams move to Xi when most of their experiment volume sits outside the codebase.

  • No SDK in the codebase. Marketing teams do not have to ask engineering to wire a flag for every test. Xi is plain-language contracts, end of story.
  • One archive across channels. A paid test, a pricing test, an outbound test, an onboarding rewrite — all are the same shape of object. GrowthBook silos by feature flag.
  • The agent runs it. Xi exposes a remote MCP server. Claude can commit experiments, log values on a cadence, and confirm ship-or-kill. GrowthBook has APIs; no native MCP today.
  • Decision by contract, not p-value. Marketing experiments often run on small N where chasing significance is worse than committing to a threshold up front. Xi codifies that decision rule.

If you are shipping product features behind flags, keep GrowthBook. If you are running marketing experiments, switch.

THE_BREAKDOWN // dimension by dimension

Where the two tools actually differ.

Dimension
Xi
GrowthBook
Unit of work
A contract (hypothesis, metric, kill threshold, end date)
A feature flag with an exposed variant
Where it runs
Outside the codebase, channel-agnostic
Inside the codebase via SDK calls and flag evaluations
Best surface
Marketing: paid, content, outbound, pricing, onboarding
Product: feature rollouts, in-app variants
Decision rule
Pre-committed kill and success thresholds, time-bound
Statistical significance via Bayesian or frequentist engine
Agent / MCP
Remote MCP server. Your agent runs the contract
SDK + REST API. No native MCP today
Free plan
Unlimited experiments, unlimited archive, no card required
Open-source self-hosted; Cloud has paid tiers
THE_HONEST_TAKE // not every tool fits every job

When each tool is the right call.

GrowthBook is the right call when
  • You ship features behind flags and need a clean stats engine to evaluate them.
  • You self-host your stack and value open-source ergonomics.
  • Your experiments are product-side: feature variants, A/B at the SDK level.
  • You have a product engineering team that owns flag rollouts and SDK integration.
Xi is the right call when
  • Your experiments are marketing decisions, not feature rollouts.
  • You want one place where every experiment lives, regardless of channel.
  • You want the agent to commit and decide experiments via MCP.
  • You want zero codebase footprint and zero SDK to maintain.
FAQ // the questions buyers actually ask

Common questions, short answers.

How is Xi different from feature flag platforms?

Feature flags are a delivery mechanism for product variants. Xi is a contract layer for marketing decisions. Different unit of work, different decision rule, different audience.

Can I use Xi alongside GrowthBook?

Yes. Use GrowthBook for in-code product experiments where flags and a stats engine are the right fit. Use Xi for the marketing experiments and the contract archive.

Is Xi open source?

Xi is hosted today, not open source. The free plan is unlimited and the contract data is portable via API.

Do I need to install an SDK?

No. Xi has zero codebase footprint. You write the contract in plain language and log values via the Claude MCP integration or manually.

Take one idea. Turn it into an experiment.

Free plan, unlimited archive, no card required. See it in Claude / Cursor / Codex in 30 seconds.