integrations tools optimizely
Optimizely MCP for your AI agent
Cogny's Optimizely integration exposes 12 tools (7 read, 4 write, 1 destructive) to an AI agent over MCP. The same tools are available on every plan, from your own AI client or inside the Cogny app.
- 7readonly reads data.
- 4writecreates or changes something.
- 1destructivedeletes, removes or replaces something.
Over MCP your own AI client asks before it runs a write. To take a tool away from your agent, block it in Settings → Integrations. Plans, approvals and limits
set it up.
Add Cogny to your AI client once, then connect Optimizely in Cogny. Every Optimizely tool below appears in the client.
- Claude CodeRun
claude mcp add --transport http cogny https://app.cogny.com/mcp, then authorise in the browser. - Claude.aiSettings → Connectors → Add custom connector, URL
https://app.cogny.com/mcp, then log in and pick your workspace. - ChatGPTSettings → Apps → Advanced settings → Developer mode → Create app, URL
https://app.cogny.com/mcp, authentication OAuth. - CursorIn Cogny, Settings → Data & Integrations → Connect AI Clients shows the block to paste into
.mcp.json, with an API key. - CodexRun
codex mcp add cogny --url https://app.cogny.com/mcp.
Notes
Optimizely connects with a personal access token from Optimizely → Profile → API Access → Generate New Token. There are no scopes: the token acts with the role its Optimizely user has on each project. A Viewer's token can read experiments and results, but creating, starting, pausing or editing an experiment needs the Editor, Project Owner or Administrator role.
Results come from Optimizely's Stats Engine as it computes them: per metric and variation, the conversion rate, visitors (samples), lift, statistical significance, confidence interval, whether the result is significant at the project's threshold, the declared winner or loser, and Optimizely's estimate of how many more visitors a result needs. Results can be filtered to a time window and segmented by one of device, browser or traffic source.
Traffic is shown and set in percent: the share of visitors who enter the experiment and the split between variations, which must add up to 100%.
create_experiment makes an A/B experiment with empty variations and doesn't start it; the variation content is designed in Optimizely's Visual Editor. start_experiment puts an experiment live on your site and is marked destructive so MCP clients ask before it runs. Traffic, audience and name changes on a running experiment are saved without going live unless publish is set. Every write stays ask-first until you allow it in Settings.
Lists return at most 100 items per call and page with page.
Known gaps
- Web Experimentation only: Feature Experimentation flags and flag rules (Optimizely's separate Flags API) and Personalization campaigns aren't covered.
- Variation content (visual changes, custom code) can't be created or edited; design variations in the Visual Editor.
- No archive or delete for experiments, and no create or edit for audiences, pages or events.
- Multivariate tests and multi-armed bandits can be read, but
create_experimentonly creates A/B tests. - Results are totals for the window; the daily timeseries and CSV export aren't exposed.
- Segmenting results by a custom attribute isn't exposed; device, browser and source are.
what each tool does.
Read 7 · only reads data
list_projectsReadList the Optimizely projects this token can see, with their platform, status and significance threshold.list_experimentsReadList a project's experiments with their status, variations, traffic split and audiences.get_experimentReadGet one experiment in full: variations and their weights, traffic, audiences, pages, metrics and schedule.get_experiment_resultsReadGet an experiment's results: conversion rate, lift, statistical significance and confidence interval per metric and variation.list_audiencesReadList a project's audiences with their targeting conditions and how many experiments use them.list_pagesReadList a project's pages (the URLs experiments run on) with their URL, type and activation.list_eventsReadList a project's events (clicks, pageviews and custom events) that experiments can measure as metrics.
Write 4 · creates or changes something
create_experimentWriteCreate a new A/B experiment in Optimizely as a draft that doesn't run until it is started.pause_experimentWritePause a running Optimizely experiment so no new visitors enter it.update_experiment_trafficWriteChange how much of the site's traffic enters an experiment and how it is split between variations.update_experimentWriteRename an Optimizely experiment or change its description (hypothesis) or audience targeting.
Destructive 1 · deletes, removes or replaces something
start_experimentDestructiveStart an Optimizely experiment, or resume a paused one, so real visitors see it.
Exact input schemas come from tools/list once Optimizely is connected; the free plan is enough to check. Missing something? Your agent can ask for it with request_feature.