Artifacts for Claude Code, Codex
and every other agent CLI
Your agent builds a dashboard, a chart or a report. Instead of leaving a file path in the terminal, it shows you the page: a block above the composer that opens in one click, and a push that opens it on your phone.
One tool for every agent CLI with an MCP client, the same on macOS, Windows and Linux. No hosted service and no account with one vendor: the page is a file on your computer, and it stays there.
view_present( path: "reports/q3.html", title: "Q3 dashboard" )
The agent calls one tool, and its page lands as a block above the composer, with one more document on the Documents tab.
Agents are good at turning a question into a page: a dashboard of your test runs, a chart of a migration, a report comparing three options. Then the page ends up as a file path somewhere in a terminal, and you have to spot it, open it and hope you picked the right one.
AgentsRoom gives every agent a tool for that last step. When the page is ready, the agent calls view_present with the path and a title. A block appears above that agent's composer. Click it: the Documents tab comes to the front and the page opens, rendered, with its charts running. On your phone, the push opens the same page.
It does not depend on the CLI. The tool lives on the AgentsRoom MCP server, which the app connects to every agent that has an MCP client, so a page from Codex behaves exactly like a page from Claude Code. Nothing leaves your computer to make it work.
What happens when an agent shows you a page
A page your agent wrote gets the same treatment whichever CLI wrote it.
The same gesture on every CLI
Claude Code, Codex, Antigravity CLI, Kimi Code and the others call the same tool and get the same result. Switching CLI does not change how you receive their work.
A block above the composer
The document lands where you are already talking to the agent, with its title and how long ago it arrived. One click opens it, the cross takes the block away and the document stays in Documents.
It does not take your screen
Nothing jumps to the front and no terminal tab appears, so the pane you are typing in stays yours. If you are working on another agent or another project, the block waits for you and a notification tells you it is there.
On your phone in one tap
When you are away from the computer, a push names the agent, the project and the page. Tap it and the page opens in the AgentsRoom app, on top of that agent's console.
Kept in Documents
Every page joins the project's Documents panel, whose tab shows how many there are. Each new presentation of a file is kept as a version, with the agent, provider and model that made it.
More than HTML
HTML pages run their own scripts, so generated charts work. The same tool also shows an image, an SVG, a PDF, a Markdown file or a CSV table, rendered rather than as source.
How to use it
Nothing to install or turn on. Four steps from the request to the page.
- 1
Ask for a page
In any agent, ask in plain words: build a dashboard of these results as one HTML file and show it to me. Asking for one file matters, because a page with its CSS, scripts and data inside it renders the same on the computer and on the phone.
- 2
The agent shows it when it is ready
Once the file is written, the agent calls view_present with its path and a short title. The title becomes the name of the block and the text of the notification.
- 3
Read it where you are
At the computer, click the block above the agent's composer: the Documents tab comes to the front and the page opens. A click on the notification does the same from anywhere in the app. Away from it, tap the push on your phone.
- 4
Ask for changes
Tell the agent what to change. It rewrites the file and shows it again: the block updates, and the earlier version stays in Documents.
How this differs from Claude's artifacts
Claude has its own artifacts. On claude.ai and in Claude Code, a page is published to Claude's service and read in the browser, with a link that people in your organization can open. That is a good fit when the point is to share the page, and it belongs to Claude: the other CLIs you run do not get it.
Here the page stays a plain file on your computer. Any CLI with an MCP client can show one, nothing is uploaded, and the file can sit in your repository next to the code that produced it. Your phone reads it through the same end-to-end encrypted connection as the rest of the app.
The two do not compete. Use Claude's artifacts when you need a link to send to a colleague, and this when you run several CLIs and want to see their work where you already are.
Limits
- There is no share link: the page is a file on your computer. To send it to someone, send the file.
- The desktop app must be running, since it holds the file and shows the block. The phone reads the page through it.
- The phone receives the page as one document, so it cannot load files next to it. On the computer, a page can use sibling images, stylesheets, scripts and data. Keep everything in one file if you want the same result on both.
- The tool reaches agents through MCP, so Aider, which has no MCP client, leaves the file path in its answer.
- A page's scripts run on your computer and can reach the network. Treat a page you did not ask for like a program you did not write. A file that looks like it holds a secret is refused.
- The personal assistant has no Documents panel: a click on its block opens the page in its own preview.
Frequently asked questions
Does Codex have artifacts like Claude?
In AgentsRoom, yes: a Codex agent can show you an HTML page it wrote with the same tool as Claude Code. The page lands as a block above the composer, opens in one click and reaches your phone. It runs on your computer, so it needs no account with a hosting service.
Is this an alternative to Claude artifacts?
For seeing an agent's work, yes, and it works with any CLI that has an MCP client. For sharing, no: Claude's artifacts give you a link for your organization, while here the page stays a file on your computer.
How do I preview an HTML file my agent wrote?
Ask the agent to show it to you: it calls view_present, and a click on the block above the composer opens the page, rendered. You can also open any HTML file yourself from the file link in the agent's answer or from the Files tab of the right panel. Scripts run, so generated charts draw.
Does my page or my data go through a server?
The page does not: it stays a file on your computer, and the phone reads it through the end-to-end encrypted connection to your desktop. The push notification goes through a push service and carries the title you see on it, the agent and project names, and the file's path inside the project.
Will it interrupt what I am doing?
No. The page never opens by itself: it waits as a block above the agent's composer until you click it. If you are not looking at that agent, you also get a notification and a push. Both follow your notification settings, and an agent announces at most one page per minute.
You may also like
Mermaid diagrams
Claude Code and Codex write Mermaid diagrams as text a terminal can only print. AgentsRoom draws them where you read the agent: chat, .md files, plans, tickets.
Conversation view
A chat view for Claude Code, Codex and five more agent CLIs: the session read from the CLI's own transcript, tool calls folded, answers live, on desktop and phone.
Plan review
Your agent proposes a plan: read it as a document, comment on, replace or delete any passage, then approve it or send it back in one message. Every version kept.
MCP Dashboard
Learn the Model Context Protocol (MCP), understand MCP servers and MCP clients, and discover how AgentsRoom becomes a visual MCP dashboard for Claude Code agents using MCP servers. Run multi-agent MCP workflows, debug MCP connections with --mcp-debug, and monitor every MCP-powered Claude Code session in real time.
See what your agents build
Download AgentsRoom for macOS, Windows or Linux, then ask any agent to build a dashboard as one HTML file and show it to you.
Companion app: monitor your agents on the go
Bring your own: Claude, Codex, Antigravity CLI, or other AI provider.
Push bugs and requests straight to your public backlog.