Overview
- MCP goes remote. Optimizely has introduced a Remote MCP Server for Experimentation, allowing for integration with browser-based AI tools such as Claude, ChatGPT, and Cursor, beyond the previous IDE-specific approach.
- No development skills required. Product managers and experimentation leads can now set up flags, configure tests, and retrieve results using straightforward language—eliminating the need for API expertise or coding.
- Preserved permissions. The server uses OAuth for authentication and maintains existing Optimizely platform permissions, ensuring users can only perform actions through MCP that they are already authorized to do in the platform’s interface.
On April 29, Optimizely launched a Remote MCP Server for Experimentation, expanding its previous IDE-based MCP integration to include browser-compatible AI tools such as Claude, ChatGPT, and Cursor. This advancement offers product managers, program managers, and experimentation teams direct engagement with the Optimizely platform through familiar AI interfaces, without requiring any coding or API knowledge.
The server supports OAuth authentication, leveraging existing Optimizely credentials and retaining platform permissions. This ensures that users can only make actions through MCP within the scope of their existing permissions in the Optimizely interface. It is equipped to handle both Web and Feature Experimentation and is presently accessible to all Optimizely Experimentation customers without the need for separate sign-ups or waitlists.
This launch aligns with a larger industry trend where MCP is emerging as the norm for linking AI agents with enterprise data and workflows. Optimizely’s strategic position at the crossroads of AI, experimentation, and content management highlights the growing demand for effective operational AI applications within digital experience platforms.
The organization maintains that this solution aims to reduce obstacles for non-technical users, thereby integrating experimentation as a fundamental component of contemporary digital operations.
Table of Contents
Understanding MCP and Its Importance for Experimentation Teams
MCP, or Model Context Protocol, is an open standard that facilitates seamless communication between AI tools and external platforms. Instead of requiring bespoke integrations for each AI client, MCP establishes a unified protocol—any compatible AI assistant can link with any MCP-enabled service.
Utilizing this protocol, Optimizely’s Remote MCP Server enables AI assistants to directly access its experimentation platform. Consequently, an AI tool transcends its role as merely a writing or coding assistant and becomes an active collaborator in experimentation, capable of creating flags, configuring tests, and retrieving results autonomously.
Functions of the Remote MCP Server
This launch broadens MCP access beyond just developers. Previously, Optimizely offered a local MCP Server solely for IDE environments. The Remote MCP Server is now hosted by Optimizely and is compatible with browser-based AI tools like Claude.ai and ChatGPT, allowing product managers, program managers, and experimentation leaders to interact with Optimizely directly without dealing with APIs or coding.
Authentication is managed via OAuth with existing Optimizely credentials—eliminating the need for API keys or local server setups. The server also inherits existing platform permissions, ensuring users can only engage in actions through MCP that they are authorized for in the Optimizely user interface.
Optimizely outlines five primary capability areas:
Overview of Optimizely MCP Capabilities
| Capability | What It Enables | Example Prompt |
|---|---|---|
| Flag and experiment management | Create and configure feature flags and A/B tests using natural language. | “Set up an A/B test for the checkout flow with 3 variations, tracking conversion on purchase_completed.” |
| Results and status queries | Retrieve experiment results, flag status, and audience details on demand. | “What experiment contains the custom JavaScript that’s causing ‘undefined is not a function’ errors?” |
| Flag lifecycle management | Identify and clean up stale or unused flags from a codebase. | “Show me unused flags in the authentication service codebase.” |
| SDK code generation | Produce production-ready integration code with error handling for specific frameworks. | “Generate React SDK integration for the recommendation_engine flag.” |
| Non-developer access | Enables PMs and program managers to directly tap into experimentation via AI tools. | “Show me the targeting conditions for the checkout_flow experiment.” |
Wider Implications: Experimentation Beyond the Development Team
The emphasis on non-developer capability is where this announcement bears significant weight for customer experience practitioners. Historically, experimentation initiatives faced bottlenecks involving engineering; tasks such as flag creation, test configuration, and results interpretation necessitated technical access or developer involvement. By integrating these processes through familiar AI tools, Optimizely illustrates that MCP can act as a bridge, facilitating access rather than merely serving as a convenience for developers.
The practical impact may depend on team structures and the level of experimentation governance needed before a flag can be created or a variant launched. Teams in more lenient environments are likely to see immediate benefits, while organizations with stricter change-control protocols may need to rethink what “inherited permissions” entail in their particular contexts before expanding access to wider teams.
Future Directions for Optimizely’s MCP
Optimizely has laid out various potential paths to explore beyond this latest launch, including autonomous campaign management agents that handle interconnected personalization efforts, proactive AI suggestions for reducing code risk via feature flags, natural language analyses of experiment performance, and integration with the Optimizely Data Platform for syncing audience segments and auto-generating targeted experiments.
The Remote MCP Server is currently accessible to all Optimizely Experimentation customers—covering both Web and Feature Experimentation—without any requirements for separate sign-ups or waitlists.
Optimizely’s Ambitious AI Strategy for Digital Experience Platforms
Over the past year, Optimizely has aggressively expanded its AI-driven digital experience platform vision, anchored by its Opal agentic AI platform. In early 2025, the company surpasses Adobe, securing the top spot in the Gartner Magic Quadrant for Digital Experience Platforms. Additionally, Gartner named Optimizely a Leader in its 2026 Magic Quadrant for Personalization Engines and has recognized it for the ninth consecutive year in the Content Marketing Platforms quadrant.
A May 2025 Opal upgrade further deepened integration across the Optimizely One suite, focusing on streamlined automation for content creation and experimentation in a SOC 2-compliant environment.
MCP & AI Experimentation: An Introduction
The Model Context Protocol is rapidly becoming the interoperability standard that links enterprise AI agents with experimentation and feature management tools. This open standard facilitates agent interactions with platform APIs—including those for feature flags, A/B testing, and personalization—through natural-language interfaces rather than requiring custom integrations.
Optimizely’s MCP Server Launch
With Optimizely’s MCP server, AI agents are capable of programmatically managing experiments. Experiment configuration, flag management, and workflow triggers are transformed into actions accessible through chat assistants and IDE tools—eliminating the need for a separate dashboard.
As previous CMSWire analysis indicated, MCP equips agents with the context necessary to navigate across various workflows—content management, CRM, analytics, and campaign planning—without the complications of unstable integrations.
Governance at the Protocol Level
Across different vendors, the MCP server format typically follows a consistent layout: a bridge that connects AI tools to platform APIs while enforcing specified read/write controls, transforming prompts into actionable platform tasks. With respect to feature flag management, this introduces permission scoping and audit capabilities at the protocol level.
Key questions for evaluation include:
- How detailed are the permissions and audit trails for agent-driven actions?
- Can agents operate across experimentation, personalization, and analytics without fragile logic?
- Which repetitive workflows can be automated safely?
- How predictable are the costs associated with agent consumption at anticipated usage levels?
The Broader Enterprise Signal
Companies like Sitecore, Contentstack, and Acquia are developing MCP-compatible access for their platforms. However, some features, particularly those from Acquia, remain on the roadmap without set release dates. This week, Acquia announced plans for interoperability with Model Context Protocol-compliant agents—including Claude, Cursor, and GitHub Copilot—with general availability scheduled for a future release.
For teams assessing MCP-centered experimentation tools, practical recommendations from earlier CMSWire reporting advise beginning with low-risk pilot projects, auditing existing dependencies in the tech stack, and establishing data-sharing boundaries prior to scaling automation driven by agents.
Have a tip to share with our editorial team? Reach out to us:
