How to Share Claude Code Output With Your Team
Four ways to get what Claude Code made in front of other people (a pull request, a Claude Code artifact, a static host, or a shared workspace), with the requirements, limits and fixes for each.
Claude Code works in a terminal, so everything it makes starts on one computer: files in a project folder, or text in a session. Sharing that work means picking a place other people can open. The right place depends on who the readers are and what they need to do with the result: review it, look at it, or keep working in it.
This guide compares the four common options, with the steps, requirements and limits of each.
Key takeaways
- Output that belongs to the codebase (scripts, docs, reports for developers) is best shared as a commit and a pull request.
- Claude Code artifacts publish a single interactive web page on claude.ai. They need a Pro, Max, Team or Enterprise plan and Claude Code signed in with your claude.ai account.
- A static host such as GitHub Pages works for HTML you want on your own address, but you manage the files and access yourself.
- When the output is a spreadsheet, document or deck that a team will edit, connect Claude Code to a shared workspace over MCP so the files live where the team works.
Option 1: Commit the output and open a pull request
If the output belongs with the code, such as a migration script, a README, a test report or an architecture note, the repository is the natural place for it. Ask Claude Code to write it to a file, commit it on a branch and open a pull request. Reviewers see it in the tool they already use, with history and comments.
This works well for developer audiences. It works less well when the readers are not in the repository, or when the output is a chart, a dashboard or a file that someone needs to open and use rather than read as a diff.
Option 2: Publish a Claude Code artifact
An artifact, in Anthropic's words, is "a live, interactive web page that Claude Code publishes from your session to a private URL on claude.ai." It suits output that is easier to see than to read: an annotated diff, a dashboard of data the session pulled, several design options side by side, or a timeline that fills in while a long task runs.
How to create and share one
- Ask for it in plain language, for example: "Make an artifact that walks through this pull request with the diff annotated inline." Claude can also publish one on its own when the output suits a page.
- Open the link. Claude publishes the page, prints its address and opens it in your browser. Until you share it, only you can see it.
- Share it from the page header. On Team and Enterprise plans you can share with specific people or everyone in your organization and make people editors. Links that anyone can open are also available; on Team and Enterprise an Owner has to allow them first.
- Keep it current. Ask Claude to revise the page and it republishes to the same address, so anyone with the page open sees the update.
Limits to know before you choose it
- One page, no backend. The documentation describes an artifact as "one self-contained page with no backend, so it can't serve multiple routes."
- Restricted outside content. The page can load fonts from Google Fonts and scripts from a short list of public code hosts. Other outside images, scripts and styles are blocked, so images are embedded in the page itself.
- Size. The finished page must be 16 MiB or smaller.
- Downloads. Offering a file download, such as a CSV or an image, depends on a capability that claude.ai turns on per account.
When publishing does not work
If Claude Code saves a local web page instead of giving you a link, artifacts are not available in that session. Anthropic's documentation lists the requirements:
| Requirement | What must be true |
|---|---|
| Plan | Pro, Max, Team or Enterprise |
| Sign-in | Claude Code is signed in with your claude.ai account. Sessions that use an API key, a gateway or cloud-provider credentials cannot publish. |
| Model provider | Anthropic's own API, not Amazon Bedrock, Google Cloud or Microsoft Foundry |
| Organization policy | Customer-managed encryption keys, HIPAA and zero data retention are not enabled |
| Where it runs | The Claude Code terminal app or the Claude desktop app. Artifacts are off by default in scripted runs, the Agent SDK and the GitHub Action. |
An administrator or a personal setting can also turn artifacts off.
Option 3: Host the HTML yourself
If you want the page on your own address, or your plan does not include artifacts, ask Claude Code to save the output as a static HTML file and publish it with a static host. GitHub describes GitHub Pages as "a static site hosting service that takes HTML, CSS, and JavaScript files straight from a repository on GitHub."
The trade-off is control against effort: you choose the address and keep the files, but you also handle who can see the page, updating it and removing it. Check who will be able to open the site before you publish anything internal.
Option 4: Have Claude Code work in a shared workspace over MCP
The first three options share a finished result. When the output is something a team keeps working in, such as a budget spreadsheet, a brief or a slide deck, a better fit is to have Claude Code create it directly in a workspace the team already uses.
Claude Code connects to outside tools through the Model Context Protocol. The protocol's own documentation calls MCP "an open-source standard for connecting AI applications to external systems." Any workspace that runs an MCP server can receive Claude Code's work: you add the server to Claude Code once, with its address and a key, and the workspace's actions become tools Claude Code can use.
What to look for in such a workspace:
- Real file formats, so people can download and keep working in Excel, Word or PowerPoint.
- Comments on specific places (a cell, a slide, a sentence) that the agent can read and answer.
- Roles, so you can give people and agents view, comment or edit access.
- A way to share outside the team, such as a public page or a link.
For how MCP servers work and how agents connect to them, see What is an MCP server?
Which option to choose
| Your situation | Best fit |
|---|---|
| The output belongs with the code and the readers are developers | A commit and a pull request |
| You want to show a result as one interactive page, privately, to your organization or publicly | A Claude Code artifact |
| You need the page on your own domain, or artifacts are not available to you | A static host |
| The team will edit the files and keep asking the agent for changes | A shared workspace over MCP |
The options combine. A common pattern is code in the repository, an artifact to walk reviewers through a change, and the spreadsheet or deck that other teams need in a shared workspace.
Frequently asked questions
Do people need a Claude account to open a shared artifact?
For an artifact shared within your organization, viewers sign in to claude.ai as members of it. For public links, Anthropic's pages differ: the Claude Code documentation says a public link needs "no claude.ai sign-in", while the help center's sharing article says everyone needs a Claude account, except for legacy artifacts published from a chat. Open the link in a private browser window before you send it.
Can teammates comment on a Claude Code artifact?
Yes, when it is shared within your organization. People can leave comments on the page, and someone who can edit the artifact can send a comment to Claude. People who only have a public link do not see the comments.
Can I share a spreadsheet or slide deck this way?
An artifact is a web page, not an Excel or PowerPoint file. To hand over files people will edit, have Claude Code save the file itself and send it, or create it in a shared workspace over MCP where the team can edit it and download it in the original format.
Does the same work for Codex?
The artifact steps in this guide are specific to Claude Code. Options 1, 3 and 4 work with Codex as well, since Codex also connects to MCP servers. See Claude Code vs Codex for how the two compare.
