Get started

Client setup

Connect Claude Code, Cursor, Claude Desktop, ChatGPT, Codex, Grok Bot, OpenCode, Devin, Kimi Code CLI, fx, or any MCP client. Most need only the URL and a browser sign-in.

Every client connects to the same endpoint. Cursor has a one-click button, and Claude has a button that fills in its form for you. The others take a command, a config file, or a short form, and none of them needs a token: you sign in and approve in your browser.

Claude CodeLink to this section

Run one command. There is no token to copy.

claude mcp add --transport http artifacts https://artifacts.space/api/mcp

What happens next

  1. Start Claude Code, type /mcp, choose artifacts, then Authenticate. Your browser opens.
  2. Sign in to artifacts.space, then approve the connection. Turn on Read-only access if the agent should only read.
  3. The agent sees the artifacts tools. It stays signed in, and you can revoke it on the Connections page.

CursorLink to this section

Add to CursorOpens Cursor with the server filled in.

Or add it by hand

Add this to .cursor/mcp.json in your project, or to ~/.cursor/mcp.json for every project.

mcp.jsonjson
{
  "mcpServers": {
    "artifacts": {
      "url": "https://artifacts.space/api/mcp"
    }
  }
}

What happens next

  1. Cursor asks you to confirm the server. When it asks you to sign in, your browser opens.
  2. Sign in to artifacts.space, then approve the connection. Turn on Read-only access if the agent should only read.
  3. The agent sees the artifacts tools. It stays signed in, and you can revoke it on the Connections page.

The button only fills in the server URL. Nothing is installed until you confirm in Cursor, and no command runs on your computer.

Claude Desktop and claude.aiLink to this section

Add to Claude Desktop (opens in a new tab)Opens claude.ai with the name and the URL filled in.

To open the form by hand, open Customize, then Connectors. Choose + Add, then Add custom connector. Name it artifacts.space and paste the URL .

In Claude's form

  1. Check the name and the URL, then choose Continue.
  2. Keep the sign-in settings Claude finds, with Use Claude's published identity as the OAuth client. Add the connector.

What happens next

  1. Claude asks you to sign in. Your browser opens.
  2. Sign in to artifacts.space, then approve the connection. Turn on Read-only access if the agent should only read.
  3. The agent sees the artifacts tools. It stays signed in, and you can revoke it on the Connections page.

Check the connection

In Claude, open Customize, then Connectors: artifacts.space is in the list. To use it in a chat, choose the + button, then Connectors, and turn it on.

The connector lives in your Claude account, so it shows in Claude Desktop, on claude.ai, and in the mobile apps. Free plans can add one custom connector.

The button only fills in the form. Nothing is added until you confirm in Claude.

On Claude Team and Enterprise plans, an owner adds the connector first: open the form for your organization (opens in a new tab), or go to Organization settings, then Connectors. Then each person finds it in Customize, Connectors, and chooses Connect to sign in as themselves.

ChatGPTLink to this section

There's no one-click link for ChatGPT yet. It takes about a minute by hand.

  1. In ChatGPT on the web, open Plugins. Choose +, then Add custom MCP server.
  2. Name it artifacts.space. Under Connection, paste the server URL .
  3. Choose OAuth for authentication. Leave the client ID and secret empty: ChatGPT registers itself.
  4. Choose I understand and want to continue, then Create as a plugin. Then install the plugin.

What happens next

  1. ChatGPT asks you to connect. A sign-in window opens.
  2. Sign in to artifacts.space, then approve the connection. Turn on Read-only access if the agent should only read.
  3. The agent sees the artifacts tools. It stays signed in, and you can revoke it on the Connections page.

Use it in a chat

Start a new chat, type @, and choose artifacts.space. It reads without asking. Before it saves, changes, or deletes anything, it asks you. You can tell it to remember your answer for that tool until the chat ends.

On ChatGPT Pro, ChatGPT can use the read tools: it finds and reads your artifacts, comments, and files. To save or change them, you need a Business, Enterprise, or Edu workspace.

You do not need Developer mode. In a Business workspace, only owners and admins can add a custom server. In Enterprise and Edu, an admin can give members the Create plugins with MCPs permission.

ChatGPT on the web does not read the Codex config. If you added the server for Codex, add it here too. When new tools arrive, open the plugin and choose Refresh.

What ChatGPT saves shows via ChatGPT. Apps built on OpenAI's API can connect the same way and show as ChatGPT too. To fix the label for one connection, choose its agent on the Connections page.

CodexLink to this section

There's no one-click link for Codex yet. It takes about a minute by hand.

In the Codex app

  1. Open Plugins. Choose Add, then Add custom MCP server.
  2. Name it artifacts.space. Under Connection, keep Server URL and paste .
  3. Set Authentication to OAuth. Leave Advanced OAuth settings as they are.
  4. Select I understand and want to continue, then choose Create as a plugin.

In the terminal

codex mcp add artifacts --url https://artifacts.space/api/mcp

Codex adds the server, sees that it uses sign-in, and opens your browser. The Codex CLI and the IDE extension share one config, so the command covers both.

In config.toml

Or add this to ~/.codex/config.toml, then sign in with the command below.

config.tomltoml
[mcp_servers.artifacts]
url = "https://artifacts.space/api/mcp"
codex mcp login artifacts

What happens next

  1. Codex asks you to sign in. Your browser opens on the sign-in page.
  2. Sign in to artifacts.space, then approve the connection. Turn on Read-only access if the agent should only read.
  3. The agent sees the artifacts tools. It stays signed in, and you can revoke it on the Connections page.

Check the connection

codex mcp list

In the Codex app, open Plugins: artifacts.space shows under Installed. In the terminal, the list shows artifacts.

Grok BotLink to this section

There's no one-click link for Grok Bot yet. It takes about a minute by hand.

In the desktop app

  1. At the bottom of the sidebar, choose the apps button, above your picture. Connect Apps opens.
  2. Scroll to the bottom of the list and choose Add a custom connector.
  3. Name it artifacts.space and keep the type Remote HTTPS.
  4. Paste the server URL . Leave the HTTP headers and the Client ID empty.
  5. Choose Add plugin. Grok Bot asks you to sign in. If it shows Needs auth instead, choose Authenticate.

Or ask a Bot

Send this message in a chat. The Bot adds the server. If it shows a sign-in card, choose Connect.

Messagetext
Add the MCP server at https://artifacts.space/api/mcp and name it artifacts.space.

What happens next

  1. Your browser opens on the sign-in page.
  2. Sign in to artifacts.space, then approve the connection. Turn on Read-only access if the agent should only read.
  3. The agent sees the artifacts tools. It stays signed in, and you can revoke it on the Connections page.

Check the connection

Open Connect Apps. artifacts.space shows under Installed. Added means it is installed but not signed in yet. If it shows Needs auth or Disconnected, choose Authenticate.

In a team, an admin's MCP policy decides which servers you can add. If the app refuses to save the server, ask your admin to allow it.

Grok on the web takes the same URL: open grok.com/connectors, choose New Connector, then Custom.

OpenCodeLink to this section

Run one command. There is no token to copy.

opencode mcp add artifacts --url https://artifacts.space/api/mcp

The command adds the server to this project. Add --global to use it in every project.

Sign in

opencode mcp auth artifacts

What happens next

  1. Run the command above, or open OpenCode, type /mcps, and choose artifacts. Your browser opens.
  2. Sign in to artifacts.space, then approve the connection. Turn on Read-only access if the agent should only read.
  3. The agent sees the artifacts tools. It stays signed in, and you can revoke it on the Connections page.

Check the connection

opencode mcp list

The list shows artifacts as connected. If it shows needs authentication, sign in again.

In opencode.json

Or add the server to opencode.json in your project by hand. OpenCode turns on sign-in for remote servers by itself.

opencode.jsonjson
{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "servers": {
      "artifacts": {
        "type": "remote",
        "url": "https://artifacts.space/api/mcp"
      }
    }
  }
}

On a remote machine, your browser can fail to reach OpenCode after you sign in. If that happens, copy the URL from the address bar and paste it into OpenCode.

DevinLink to this section

There's no one-click link for Devin yet. It takes about a minute by hand.

  1. Open Customize, then MCPs. Choose Add MCP, then Add custom MCP.
  2. Name it artifacts and choose the HTTP transport.
  3. Paste the server URL and choose OAuth as the authentication method. Set Access to Personal.
  4. Save, then choose Connect next to the server.

What happens next

  1. Devin gives you a link. Open it to sign in.
  2. Sign in to artifacts.space, then approve the connection. Turn on Read-only access if the agent should only read.
  3. The agent sees the artifacts tools. It stays signed in, and you can revoke it on the Connections page.

Adding a custom MCP server needs the Manage MCP Servers permission in Devin. Without it, use Suggest MCP Integration to ask an admin.

For a Devin that works for your whole team, add it as an agent teammate instead. Choose Auth Header, keep the key Authorization, and set the value to Bearer and the agent's token. Its work then shows under its own handle.

Kimi Code CLILink to this section

There's no one-click link for Kimi Code CLI yet. It takes about a minute by hand.

Add this to ~/.kimi-code/mcp.json for every project, or to .kimi-code/mcp.json in one project. Or type /mcp-config in Kimi Code and ask it to add the server.

mcp.jsonjson
{
  "mcpServers": {
    "artifacts": {
      "url": "https://artifacts.space/api/mcp"
    }
  }
}

Sign in

Start a new Kimi Code session, because open sessions do not see a new server. Then type:

Kimi Codetext
/mcp-config login artifacts

What happens next

  1. Kimi Code shows a sign-in link. Open it in your browser.
  2. Sign in to artifacts.space, then approve the connection. Turn on Read-only access if the agent should only read.
  3. The agent sees the artifacts tools. It stays signed in, and you can revoke it on the Connections page.

Check the connection

Kimi Codetext
/mcp

Kimi Code lists artifacts with its connection status.

Kimi Code CLI replaced the older Kimi CLI. It has no kimi mcp add command. Run kimi migrate to move your servers, then sign in to each one again.

fxLink to this section

Install fx

fx is a minimal, open-source coding agent CLI from Vercel Labs. It runs on macOS and Linux.

curl -fsSL https://fx.sh/setup.sh | bash

Connect it

Run one command. There is no token to copy.

fx mcp add artifacts https://artifacts.space/api/mcp

fx saves the server in ~/.fx/mcp.json, sees that it uses sign-in, and opens your browser. Inside fx, does the same.

What happens next

  1. Your browser opens on the sign-in page. To sign in later or again, run fx mcp login artifacts.
  2. Sign in to artifacts.space, then approve the connection. Turn on Read-only access if the agent should only read.
  3. The agent sees the artifacts tools. It stays signed in, and you can revoke it on the Connections page.

Add the skill

Type this inside fx. It installs the skill to ~/.fx/skills, so it works in every project.

fxtext
/skills install artifacts-space/skills

Check the connection

fx mcp show artifacts

fx connects and lists the artifacts tools. Inside fx, /mcp shows the same.

In mcp.json

Or add the server to ~/.fx/mcp.json by hand, then sign in with fx mcp login artifacts. In a project's .mcp.json, fx waits until you approve the server with fx mcp approve artifacts.

mcp.jsonjson
{
  "mcpServers": {
    "artifacts": {
      "type": "http",
      "url": "https://artifacts.space/api/mcp"
    }
  }
}

On a machine without a browser, such as over SSH, run fx mcp login --no-browser artifacts. fx prints a link to open on any device, and you paste back the address your browser ends up on.

Other clientsLink to this section

Point any client that supports OAuth for remote servers at over Streamable HTTP. It needs only the URL.

What happens next

  1. The client opens your browser when it first connects.
  2. Sign in to artifacts.space, then approve the connection. Turn on Read-only access if the agent should only read.
  3. The agent sees the artifacts tools. It stays signed in, and you can revoke it on the Connections page.

A request without a token gets a 401 with a WWW-Authenticate header that names the protected-resource metadata, so OAuth-aware clients find the sign-in server by themselves. Clients can register with a Client ID Metadata Document or dynamic client registration. If your client cannot do OAuth, use a token.

Headless and CILink to this section

Where nobody can approve in a browser, such as CI, a container, or a cloud agent, send an access token. Create one on the Connections page and keep it in an environment variable such as ARTIFACTS_SPACE_TOKEN, never in source control.

mcp.jsonjson
{
  "mcpServers": {
    "artifacts": {
      "url": "https://artifacts.space/api/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_TOKEN"
      }
    }
  }
}

Claude Code, reading the token from the environment:

claude mcp add --transport http artifacts https://artifacts.space/api/mcp --header "Authorization: Bearer $ARTIFACTS_SPACE_TOKEN"

Codex, in ~/.codex/config.toml:

config.tomltoml
[mcp_servers.artifacts]
url = "https://artifacts.space/api/mcp"
bearer_token_env_var = "ARTIFACTS_SPACE_TOKEN"

Give team agents their own token

An agent that works for a team should be an agent teammate with its own token, so its work shows under its own handle. See Agent tokens.

Make it the defaultLink to this section

Some agents have their own way to make pages and docs, such as the Artifact tool in Claude Code. Add one line to your agent's instructions, and it saves to artifacts.space every time. You can still ask for the built-in tool by name.

Add this linetext
Use artifacts.space (the `artifacts` MCP server) for every artifact, doc, plan, report, or page. Do not use any built-in artifact or document tool unless I name it.
AgentEvery projectOne project
Claude Code~/.claude/CLAUDE.mdCLAUDE.md
Codex~/.codex/AGENTS.mdAGENTS.md
CursorCursor Settings, Rules, User RulesAGENTS.md, or a .cursor/rules/*.mdc rule with alwaysApply: true
OpenCode~/.config/opencode/AGENTS.mdAGENTS.md
fx~/.fx/AGENTS.mdAGENTS.md
Kimi Code CLI~/.kimi-code/AGENTS.mdAGENTS.md
Devin CLI~/.config/devin/AGENTS.mdAGENTS.md
Claude Desktop and claude.aiSettings, Instructions for ClaudeThe project's instructions
Other agentsThe instructions file its docs nameAGENTS.md, which most agents read

In Cursor, a project rule needs alwaysApply: true so it loads in every chat:

.cursor/rules/artifacts-space.mdcmarkdown
---
description: Save artifacts to artifacts.space
alwaysApply: true
---

Use artifacts.space (the `artifacts` MCP server) for every artifact, doc, plan, report, or page. Do not use any built-in artifact or document tool unless I name it.

To turn off Claude Code's own Artifact tool as well, add this rule to ~/.claude/settings.json, next to the settings already there. If the file has a permissions.deny list, add "Artifact" to it. The built-in tool stays off until you remove the rule.

Add to ~/.claude/settings.jsonjson
{
  "permissions": {
    "deny": [
      "Artifact"
    ]
  }
}

Turning off Artifacts in Claude Code's /config does the same: it writes "enableArtifact": false to that file.

Product names and logos belong to their owners. They show compatibility only.