Integration · Optimizely

Web Experimentation and Feature Experimentation in one MCP.

Optimizely is the long-time enterprise leader in personalization and experimentation. The Remote MCP Server makes both products accessible from any AI client without a local install.

Browser-based ChatGPT and Claude.ai work via OAuth; the agent inherits the user's Optimizely UI permissions, so audit trails and access controls stay intact. Available to all paying customers.

Official MCPHosted Remote MCP. OAuth via existing Optimizely login.

What you can do via MCP

Example prompts the agent runs.

  • “Create a feature flag pricing-redesign with control and treatment variants.”

    Creates the flag, sets the two variants, and surfaces the SDK integration snippet for the chosen language.

  • “Set up an A/B test for the checkout flow with 3 variations, tracking conversion.”

    Creates the experiment with the right targeting and primary metric, returns the experiment ID and the dashboard link.

  • “What audiences are being used in running experiments?”

    Pulls active experiments, dedupes audiences across them, and returns a usage matrix the team can audit.

  • “Show me unused flags in the authentication codebase.”

    Cross-references flags with codebase grep, surfaces flags with no live evaluations, and proposes a cleanup PR.

  • “Generate React SDK integration code for the recommendation_engine flag.”

    Returns the React snippet wired to the flag key, with the right SDK initialization and provider setup.

Optimizely · MCP

Creating an A/B test from a brief. The Remote MCP runs hosted; the agent calls it via OAuth from the user's existing Optimizely login.

Optimizely MCP
docs
# Optimizely Remote MCP, hosted, OAuth
# Connect: ChatGPT/Claude.ai > Add Connector > Optimizely > Sign in

mcp.optimizely.create_experiment({
  type: "ab_test",
  name: "checkout-flow-3-variant",
  page_id: 8472614,
  primary_metric: "conversion",
  audiences: ["us_visitors"],
  variations: [
    { name: "Control", weight: 33 },
    { name: "Treatment A", weight: 33 },
    { name: "Treatment B", weight: 34 }
  ]
})

# Returns experiment_id and dashboard URL
One command sample showing how the agent talks to Optimizely. The MCP exposes the platform's primitives as tools; the agent translates the prompt into the right call.

MCP integration

Optimizely MCP server.

Server
Optimizely-hosted Remote MCP (added via AI client connector menu)
Auth
OAuth via existing Optimizely login
Hosting
Hosted by Optimizely; no local install required
  • Works in browser-based ChatGPT and Claude.ai with no local setup
  • Inherits the user's UI permissions, so audit trails stay intact
  • Covers Web Experimentation and Feature Experimentation in one connection
  • Available to all paying Optimizely customers

Optimizely developer docs

Visual demonstration

What this looks like in practice.

Optimizely/Experiment results
Ship treatment

checkout-flow-3-variant

Primary metric: conversion

VariantMetricLift95% CIp
Controlcontrol8.20%baseline——
Treatment A8.78%+7.07%[2.10, 12.00]0.012
Treatment B8.35%+1.83%[-3.10, 6.70]0.466
Running
18d
N
98,400

Treatment A wins at 95%. Treatment B not significant. Recommendation: ship Treatment A; archive Treatment B.

Three-variant Optimizely result via the MCP. Treatment A clears significance; Treatment B does not. The agent recommends accordingly.
Optimizely/Feature flag
Experiment

recommendation_engine

New ML-driven recommendation engine for product detail pages

Environment

production

On50% rollout

Targeting rules

  • audience IN ('us_visitors', 'returning_users')
  • rollout 50%
An Optimizely Feature Experimentation flag managed through the MCP. Environment, audience, and rollout all editable via the agent.

CLI alternative

MCP only.

Optimizely does not ship a first-party CLI for AI workflows. The Remote MCP plus the REST API cover all programmatic surfaces. Because the MCP is hosted, the team does not need to manage CLI auth tokens or install packages.

Pairs with these skills

The PM-experimentation skill suite.

This integration pairs with the forthcoming experiment-design, feature-flagging skills. The skill landing pages and SKILL.md sources land in subsequent dispatches; cross-link hyperlinks are added when the skill pages ship.

Continue reading.