Skip to content

Getting started

This guide walks through how to create an experimental environment and publishing it to a shareable link in three steps.

Prerequisites:


Terminal window
claude plugin marketplace add git@git.autodesk.com:iaas/experimental-environments.git --scope user
claude plugin install adsk-experiment@adsk-experiment --scope user

This installs everything Claude needs to create an experimental environment. Make sure --scope user is on both commands.

The adsk-experiment CLI is a separate binary. The plugin installs a hook in Claude for that: the next time you run Claude, it will install it. To install it up front instead, see install.md.

Once you’ve run Claude and installed the binary, you can connect with it directly using the Claude ! (shell) command:

Terminal window
! adsk-experiment login # opens your browser, usually without a click
! adsk-experiment doctor # signed in, control plane reachable, CLI current, Harness MCP accessible

You can use those to sign in yourself. Claude signs you in to the control plane when it first needs it.


2. Authenticate the Jenkins and Harness MCPs

Section titled “2. Authenticate the Jenkins and Harness MCPs”

Run /mcp inside Claude Code, or:

Terminal window
claude mcp login plugin:adsk-experiment:ee-jenkins
claude mcp login plugin:adsk-experiment:harness

When authenticating the Harness MCP, you will need to manually enter your Autodesk email address and enter an OTP (one-time password) that will be emailed to you.

Make sure to select the production environment (prod) when authenticating, if you have multiple environments to choose from, like Sandbox.

If the Harness MCP shows access denied, request access to the EXPENV - Development - Commercial - Viewer package, and once it’s granted, retry /mcp in Claude.

If the Harness MCP still shows “needs authentication” after you’ve entered the OTP and authenticated or you see the error “Got new credentials, but harness rejected them on reconnect”, then reach out to us on Slack in #experimental-environments.


Set me up an experimental environment for a pricing page mockup, and publish the starter site into it.

Claude creates the environment and a new GitHub repository, commits your content, then builds and deploys. Claude will suggest an experiment name and industry for you to use.

Industries: Once set, industry cannot be changed.

CodeIndustry
conConstruction
mfgManufacturing
medMedia & Entertainment
opsOperations
pltPlatform
entEnterprise

Note: When working with experimental environments, “experiment”, “environment”, and “prototype” all mean the same thing.

Prompts for Claude:

PromptWhat you get
“Publish this to an experimental environment”A link you can send to a colleague
“Give me the link to this experiment”The hostname, once it is serving
“Publish this again”The same hostname, updated with your latest commit
“Show me my experiments”Everything live, when each expires, how many slots are left
“When does this experiment expire?”The date, and how to extend it
“Extend this experiment”A new expiry, counted from now
“Why did this fail?”The failing stage and its logs

Once Claude returns with the live link, open the result in a browser holding an Autodesk SSO session:

https://<component>-<environment>.<industry>.experiment.autodesk.com

If something isn’t working, start with adsk-experiment doctor. If you are running into experience issues, visit Limits and rough edges to see if they are expected.

For what’s built, what’s in progress, or why a decision was made the way it was, see architecture/roadmap.md, architecture/overview.md, and the ADRs.

For additional help, join #experimental-environments on Slack.