---
title: Optimizely Experimentation MCP server quickstart
slug: experimentation-strategy/docs/optimizely-experimentation-mcp-server-quickstart
docTags: 
createdAt: 2026-08-05T20:33:25.228Z
---

- Optimizely Web Experimentation
- Optimizely Personalization
- Optimizely Feature Experimentation
- Optimizely Performance Edge

Get your agent connected to Optimizely Experimentation using the following steps.

## Prerequisites

- An Optimizely account with Optimizely Agent Platform enabled, connected to at least one Feature Experimentation or Web Experimentation instance.
- One of the following AI clients installed: 
  - Claude Code (CLI)
  - Claude Desktop
  - Claude.ai
  - Cursor
  - Visual Studio Code
  - Windsurf

## Add the MCP server to your AI client

Add the Optimizely MCP server URL to your client's MCP configuration:

```auto
https://exp.mcp.opal.optimizely.com/mcp
```

Where to add this depends on your client.

- **Claude Code** – Run `claude mcp add --transport http optimizely-exp https://exp.mcp.opal.optimizely.com/mcp`.
- **Claude Desktop **– Go to **Settings > MCP Servers > Add Server**.
- **Claude.ai** – Go to **Settings > Integrations > Add MCP Server**.
- **Cursor** – Go to **Settings > MCP > Add new MCP server**.
- **Visual Studio Code** – Open MCP settings in the Copilot configuration.
- **Windsurf** – Go to **Settings > MCP Servers**, or edit `.windsurf/mcp.json` in your project root.

For detailed, client-specific configuration instructions, see [Install Optimizely Experimentation MCP server](docId\:oDoyHBD4Pfoypf3Vm_G6K).

## Authenticate with Optimizely

When your AI client connects to the MCP server for the first time, it opens an Open Authorization (OAuth) flow in your browser.

1. Log in with your [Opti ID](docId\:pbehd96P529DqHZ0HOQlm) credentials (Optimizely's unified identity system, the same login you use for the Optimizely UI).
2. Connect to your Agent Platform instance.
3. Authorize the MCP connection.

You do not need to create a separate API token or manage credentials manually. The MCP server uses your Optimizely identity and respects your existing account permissions.

Your authentication session is time-limited. When it expires, your AI client notifies you that re-authentication is required. The exact prompt varies by client, but you can complete the OAuth flow again by following the prompt.

## Run your first query

After you authenticate, your agent has access to your Optimizely data. To run your first query, enter *List my Optimizely projects*.

Your agent calls the MCP server, retrieves your project list, and returns the results: project names, IDs, and status.

## Verify it works

A working connection lets you

- See Optimizely tools listed in your AI client's available tools (look for tools prefixed with `exp_`).
- Get a response to *List my Optimizely projects* that includes real project data from your account.
- Ask follow-up questions about the returned data without re-authenticating.

If the server is not responding or you do not see the `exp_` tools listed, see the Troubleshoot section of the [Install Optimizely Experimentation MCP server](docId\:oDoyHBD4Pfoypf3Vm_G6K) article.

## What to try next

When you have confirmed the connection works, try these prompts to explore the full range of available capabilities. Replace YOUR\_PROJECT\_NAME with the name of one of your Optimizely projects (visible in the Optimizely UI under **Projects**).

- **Query your experiments** – *Show me all running experiments in YOUR\_PROJECT\_NAME.*
- **Check experiment results** –* What are the results for experiment EXPERIMENT\_NAME?*
- **Get SDK implementation help** –* Help me get started with the React SDK for feature flags.*
- **Compare environments** – *What flags are enabled in production but not in staging?*

## Next steps

- [Install Optimizely Experimentation MCP server](docId\:oDoyHBD4Pfoypf3Vm_G6K) – Detailed configuration instructions for each supported AI client.
- [Optimizely Experimentation MCP server example prompts](docId\:OTSYSVVYywrW5xfsmtqGA) – A library of useful queries organized by workflow.