Integration · Kameleoon

Winning experiment to production code.

Kameleoon's MCP focuses on the implementation workflow. Pull raw variation code (JS or CSS) from a winning experiment and convert it to a production-ready React component the team ships behind a flag.

Seven specialized tools. The MCP is scoped to the convert-and-promote flow rather than broad CRUD, which is a deliberate design choice: agents work best with a small, sharp toolset for a specific job, and Kameleoon already has a strong dashboard for the rest.

Official MCPHosted MCP at mcp.kameleoon.com/mcp. OAuth via mcp-remote proxy on port 35535.

What you can do via MCP

Example prompts the agent runs.

  • “List my Kameleoon feature flags.”

    Returns the flag list scoped to the user's Kameleoon site, including current state and last evaluation timestamp.

  • “Show me the winner data for experiment 149640.”

    Pulls experiment results, identifies the winning variation, and surfaces the lift, confidence, and segment-level effects.

  • “Pull the variation code for experiment 149640 and convert it to React components.”

    Reads the raw JS/CSS variation, refactors it into a typed React component with proper props, and proposes the integration site in the codebase.

  • “Create a feature flag 'pbx-experiment-373001-winner' and gate the converted code behind it.”

    Creates the flag, wires the converted component to evaluate it, and stages a PR that ships the new code path off by default.

  • “Enable the new-search flag in production.”

    Toggles the production environment for the named flag and confirms the change took effect.

Kameleoon · MCP

The convert-winning-variation flow. The agent reads the raw variation code from Kameleoon, then proposes the React component. The same prompt also stages the flag for the rollout.

Kameleoon MCP
docs
# Kameleoon MCP, hosted, OAuth via mcp-remote proxy
# Connect: https://mcp.kameleoon.com/mcp

mcp.kameleoon.get_variation_code({
  experiment_id: 149640,
  variation_id: 373001
})
// Returns raw JS + CSS

mcp.kameleoon.create_feature_flag({
  key: "pbx-experiment-373001-winner",
  description: "Promoted from winning variation"
})
// Agent then refactors variation code into a React
// component gated by the new flag, opens PR.
One command sample showing how the agent talks to Kameleoon. The MCP exposes the platform's primitives as tools; the agent translates the prompt into the right call.

MCP integration

Kameleoon MCP server.

Server
https://mcp.kameleoon.com/mcp
Auth
OAuth via mcp-remote proxy (port 35535)
Hosting
Hosted by Kameleoon
  • Seven specialized tools focused on the convert-and-promote workflow
  • Reads winning variation JS/CSS for refactoring into production components
  • Creates flags scoped to the experiment that produced the variation
  • Pairs with the Kameleoon dashboard for broad CRUD work the MCP does not cover

Kameleoon MCP docs

Visual demonstration

What this looks like in practice.

Kameleoon/Experiment results
Ship treatment

experiment-149640 / homepage-hero-variant

Primary metric: signup_completed

VariantMetricLift95% CIp
Controlcontrol6.10%baseline——
Variation 3730017.04%+15.41%[7.80, 23.00]0.001
Running
14d
N
48,200

Significant winner. Agent's next step: pull variation_id 373001 code and convert to a React component gated behind a new flag.

A winning Kameleoon experiment. The MCP returns this view, then the convert-to-React workflow takes over for the rollout.

CLI alternative

MCP only.

Kameleoon does not ship a first-party CLI. The hosted MCP plus the standard Kameleoon REST API cover programmatic access; for AI workflows the MCP is the right interface, and the dashboard handles the broader CRUD work the seven-tool MCP intentionally does not cover.

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.