News
Research Report
Business Leaders CX Report: What’s Working and What’s Not in AI-Driven CX
The Gist
- MCP goes remote. Optimizely launched a Remote MCP Server for Experimentation, extending its earlier IDE-based integration to browser-based AI tools including Claude, ChatGPT, and Cursor.
- No dev required. Product managers and experimentation leads can now create flags, configure tests, and pull results in plain language — no API knowledge or code editor needed.
- Permissions intact. The server authenticates via OAuth and inherits existing Optimizely platform permissions, so users can only do through MCP what they're already authorized to do in the UI.
Optimizely on April 29 launched a Remote MCP Server for Experimentation, extending its earlier IDE-based MCP integration to browser-based AI tools including Claude, ChatGPT and Cursor. The move gives product managers, program managers, and experimentation teams direct access to Optimizely's platform through the AI tools they already use — no API knowledge or code editor required.
The server authenticates via OAuth using existing Optimizely credentials and inherits platform permissions, meaning users can only execute through MCP what they're already authorized to do in the Optimizely UI. It supports both Web Experimentation and Feature Experimentation and is available now to all Optimizely Experimentation customers with no separate sign-up or waitlist.
The launch reflects a broader industry trend in which MCP is becoming a standard for connecting AI agents with enterprise data and workflows. Optimizely's positioning at the intersection of AI, experimentation and content management is reinforced by this development, as demand for practical, operational AI applications continues to rise in the digital experience platform space.
The company asserts that the solution is designed to lower barriers for non-technical users, making experimentation a more integrated part of modern digital operations.Table of Contents
- What Is MCP — and Why Does It Matter for Experimentation Teams?
- What the Remote MCP Server Actually Does
- The Broader Implication: Experimentation Beyond the Dev Team
- What's Next on Optimizely's MCP Roadmap?
- Optimizely's AI Play for DXP
- MCP & AI Experimentation: A Primer
What Is MCP — and Why Does It Matter for Experimentation Teams?
MCP stands for Model Context Protocol, an open standard that allows AI tools to communicate with external platforms and services. Rather than requiring custom integrations for each AI client, MCP provides a shared protocol — any compatible AI assistant can connect to any MCP-enabled service.
Optimizely's Remote MCP Server uses that protocol to give AI assistants direct access to its experimentation platform. The practical effect: an AI tool stops being a writing or coding aide and becomes an active experimentation collaborator — one that can create flags, configure tests, and pull results on command.
What the Remote MCP Server Actually Does
The launch extends MCP access beyond developers. Earlier, Optimizely offered a local MCP Server scoped to IDE environments. The Remote MCP Server is hosted by Optimizely and works with browser-based AI tools — including Claude.ai and ChatGPT — meaning product managers, program managers and experimentation leads can interact with Optimizely directly, without touching an API or a code editor.
Authentication runs through OAuth using existing Optimizely credentials — no API keys, no local server setup. The server also inherits platform permissions, so users can only execute through MCP what they're already authorized to do in the Optimizely UI.
Optimizely describes five core capability areas:
Capability Overview for Optimizely MCP
| Capability | What It Enables | Example Prompt |
|---|---|---|
| Flag and experiment management | Create and configure feature flags and A/B tests in natural language | "Set up an A/B test for the checkout flow with 3 variations, tracking conversion on purchase_completed" |
| Results and status queries | Pull experiment results, flag status, and audience details on demand | "What experiment contains the custom JavaScript that's throwing 'undefined is not a function' errors?" |
| Flag lifecycle management | Identify and clean up stale or unused flags across a codebase | "Show me unused flags in the authentication service codebase" |
| SDK code generation | Generate production-ready integration code with error handling for specific frameworks | "Generate React SDK integration for the recommendation_engine flag" |
| Non-developer access | Gives PMs and program managers direct experimentation access through AI tools | "Show me the targeting conditions for the checkout_flow experiment" |
The Broader Implication: Experimentation Beyond the Dev Team
The non-developer angle is where this announcement carries the most weight for CX practitioners. Experimentation programs have historically bottlenecked at engineering — flag creation, test configuration and results interpretation all required technical access or developer involvement. By routing those workflows through AI tools that non-technical team members already use, Optimizely is making a case that MCP can function as an access layer, not just a developer convenience.
Whether that plays out in practice depends on how teams are actually structured — and how much experimentation governance their organizations require before someone can spin up a flag or launch a variant. More permissive environments will benefit immediately; organizations with stricter change-control processes may need to think through what "inherited permissions" means in that context before opening this to broader teams.
What's Next on Optimizely's MCP Roadmap?
Optimizely outlined several directions it's exploring beyond the current launch: autonomous campaign management agents handling interconnected personalization programs; proactive AI suggestions for where feature flags could reduce code risk; natural language experiment performance analysis; and integration with the Optimizely Data Platform to sync audience segments and auto-generate targeted experiments.
The Remote MCP Server is available now to all Optimizely Experimentation customers — both Web and Feature Experimentation — with no separate sign-up or waitlist.
Optimizely's AI Play for DXP
Optimizely has aggressively built out its AI-driven digital experience platform strategy over the past year, anchored by its Opal agentic AI platform. In early 2025, the company displaced Adobe atop the Gartner Magic Quadrant for Digital Experience Platforms. Gartner also named Optimizely a Leader in its 2026 Magic Quadrant for Personalization Engines and recognized it for the ninth consecutive year in its Content Marketing Platforms quadrant.
A May 2025 Opal upgrade deepened integration across the Optimizely One suite, targeting scalable automation for content creation and experimentation within a SOC 2-compliant environment.
MCP & AI Experimentation: A Primer
Model Context Protocol is fast becoming the interoperability layer connecting enterprise AI agents to experimentation and feature management tools. The open standard lets agents interact with platform APIs — including those for feature flags, A/B testing and personalization — through natural-language interfaces rather than custom integrations.
Optimizely's MCP Server Launch
Optimizely's MCP server lets AI agents manage experiments programmatically. Experiment configuration, flag management and workflow triggers become actions accessible through chat assistants and IDE tools — no separate dashboard required.
As earlier CMSWire analysis noted, MCP gives agents context to operate across content management, CRM, analytics and campaign workflows without brittle integrations.
Governance at the Protocol Layer
Across vendors, the MCP server pattern follows a consistent structure: an access layer connects AI tools to platform APIs with defined read/write controls, converting prompts into platform actions. For feature flag management, this introduces permission scoping and audit capability at the protocol level.
Key evaluation questions:
- How granular are permissions and audit trails for agent-driven actions?
- Can agents span experimentation, personalization and analytics without fragile logic?
- What share of repetitive workflows can be automated safely?
- How predictable are agent consumption costs at projected usage volumes?
Webinar
May
12
How Marketing Teams Can Take Control of the Web Experience
See how Veo and Kvalifik put marketing in the driver's seat, shipping faster and growing qualified bookings by 150%.
Register
Webinar
May
14
Two Audiences, One WordPress. Is Yours Ready?
The practical steps to making your WordPress portfolio ready for both human visitors and AI agents.
Register
Webinar
May
19
From Content Sprawl to Competitive Advantage: A Knowledge Operations Roadmap
How automation, AI and a centralized content platform turn knowledge fragmentation into a scalable competitive advantage.
Register
Webinar
May
20
From AI Investment to CX Results: What Enterprise Leaders Need to Know
Move beyond experiments. See how top enterprises scale AI for CX results.
Register
Webinar
On demand
Where Contact Center AI Is Actually Headed
A straight read on where the contact center platform market is going and what it means for your next technology decision.
Watch Now
Webinar
On demand
AI for Your DXP: Connect What You Have, Transform How You Work
Most AI strategies stop at the platform—but work happens elsewhere. Bring intelligence into Teams, email, tickets and CRM.
Watch Now
The Broader Enterprise Signal
Vendors including Sitecore, Contentstack and Acquia are building MCP-compatible agent access into their platforms — though some capabilities, including Acquia's, remain on the roadmap without confirmed release dates. The company this week announced planned interoperability with Model Context Protocol-compatible agents — including Claude, Cursor and GitHub Copilot — with general availability planned for a future release.
For teams evaluating MCP-based experimentation tooling, practical advice from earlier CMSWire reporting holds: start with low-risk pilots, audit existing stack dependencies and establish data-sharing guardrails before scaling agent-driven automation.
Have a tip to share with our editorial team? Drop us a line: