From zero to your AI's first answer, without signing up: connect, check, and ask
https://iceshore.ai/en/blog/a-p5-2/
On this page
This is for you if you use Claude Code or Cursor every day and want to see what comes back before you try IceShore. We walk through everything up to the first answer, with no signup and no login. You do three things by hand.
What IceShore provides is architectures and the use cases written in them (what each system does for the business). The one answering your questions is the AI you already use.
Connect with claude mcp add
This step runs in your terminal. On the connection page, choose your AI and it shows the command to paste. For Claude Code, it is one line.
claude mcp add -s user iceshore --transport http https://mcp.iceshore.ai/mcpWhen you run it, you get this back.
Added HTTP MCP server iceshore with URL: https://mcp.iceshore.ai/mcp to user config
File modified: /Users/<you>/.claude.jsoniceshore is the name your AI uses for this connection. OpenAI Codex also takes a single command, and Claude Desktop only needs the URL pasted in. For Cursor, Cline, GitHub Copilot, and Windsurf, you add a few lines to a config file. The page also shows the path to paste them into.
Note that -s user changes the settings for your whole user account. If your policies require a separate environment for each project, use a project-level scope instead.
Next, the connection page tells you to restart your AI and then start authentication. You can skip that here. Authentication is needed to use your own workspace, not to read the sample workspace. In this article, you see one answer before authenticating, and then decide whether to continue.
Check the connection with claude mcp list
You type this in your terminal as well.
claude mcp listiceshore: https://mcp.iceshore.ai/mcp (HTTP) - ✔ ConnectedIf you see this line, the connection works. You can also ask your AI, “Check whether iceshore is available.”
Note that this output may differ depending on the version of your AI client.
Ask your AI
From here on, you talk to your AI. No more commands.
Show me the sample architecture. If we stopped the medical-records database, who would be affected?
Your AI fetches the list of sample workspaces, reads the architecture, and traces the paths back to answer. The answer comes from the elements shown in the architecture. For this database, it returns four: doctors viewing medical records, patients booking appointments online, front-desk staff managing appointments, and a regulatory-compliance backup that runs every night at 2 a.m.
So far, you have not signed up or logged in. The sample workspace holds several architectures, and you can read the contents of each one.
https://app.iceshore.ai/workspace/sample/
How far you can go without signing up
| What you can do | Signup |
|---|---|
| Fetch the list of sample workspaces | Not needed |
| Read a sample workspace architecture in full | Not needed |
| Read the writing guide | Not needed |
| Fetch the list of your own workspaces | Needed |
| Create or save an architecture | Needed |
| Upload files and import them into an architecture | Needed |
When an operation is not available without signing up, your AI shows you how to sign up.
When you bring in your own architecture
Once you have a feel for it with the sample, the next step is your own system. Ask your AI to gather the material and it starts by looking for configuration-management files such as CloudFormation or Terraform (it is fine if you have none). Your AI reads them, organizes the use cases, and drafts the architecture.
There is not much for you to decide at this stage. Start by looking over the architecture it produces.
If the architecture is not what you expected, that may be good news: it shows how well your AI understands your system. From there, keep the architecture in front of you and correct it, either by talking with your AI or in IceShore's editor. That is how your AI's understanding improves.
Three places people get stuck
On the connection page, you can leave the connection name field empty. It defaults to iceshore.
Right after connecting, your AI may not recognize IceShore's tools. Some clients need a restart. If claude mcp list returns ✔ Connected for the IceShore line, the connection is there. The AI just has not reloaded yet.
Nothing you try in the sample is saved. Anonymous users can read only the sample workspace architectures, so there is nowhere to write. If you want to keep something, ask your AI to export the architecture and save it to a file. It is text in ADL (Architecture Description Language) format, so the file is a deliverable on its own.
What these three steps tell you
You have now tried the command that connects to IceShore, the command that checks the connection, and a simple question to your AI. What this shows is that your AI reads the system architecture and then starts talking about the business, not only about technical terms. With IceShore, you can get your AI to understand how your systems and your business relate.