Connect Claude, Cursor, VS Code and Codex to Dataspheres AI | DATASPHERES AI - Dataspheres AI

One hosted address, one sign-in: how to connect Claude, Claude Code, Cursor, VS Code with Copilot, and OpenAI Codex to your dataspheres, what arrives (tools, skills, your datasphere's brief), and how to check it works.

Dataspheres AI runs a hosted connection for AI assistants at https://dataspheres.ai/mcp . It speaks the Model Context Protocol, the open standard that Claude, Cursor, VS Code, GitHub Copilot and OpenAI Codex all use to reach outside tools. Once an assistant is pointed at that address and you have signed in with your Dataspheres account, it can read and create pages, decks, narrated videos, research runs, datasets, surveys, newsletters, courses and planner tasks in the dataspheres you belong to, with the same access your role gives you in the app. There is no API key to copy and nothing to install on your machine. The setup for each assistant is a short paste or a single click, described below. What arrives once you are connected The connection carries three things. The first is the tool set: a curated core of about forty tools is listed directly, and search_tools plus call_tool reach the rest of the catalogue, which is over three hundred tools wide. The second is the skills. Each of the fifty-six Dataspheres skills (pages, presentations, surveys, newsletters, spec-driven development and so on) is served as a prompt, so it shows up in the assistant's slash or @ menu, and as a readable resource at dataspheres://skills/<name> . Using a skill before working in its area gives the assistant the same house rules a Dataspheres engineer would follow. The third is your context: the first tool, get_context , returns who you are, your dataspheres with their roles, and each datasphere's purpose and standing instructions, which the assistant is told to treat as the owner's brief for that space. Claude on the web and Claude Desktop Claude adds outside tools through connectors. Open Settings → Connectors , choose Add custom connector , give it the name Dataspheres and paste the address https://dataspheres.ai/mcp . Save it, then click Connect : a Dataspheres sign-in window opens, you approve the access, and the connector shows as connected. In any chat, turn it on from the tools menu. An organisation administrator can add the same connector once for every member. The skills appear under the connector's prompts. Claude Code In a terminal, one line registers the server, and Claude Code opens the sign-in itself the first time it is used: claude mcp add --transport http dataspheres https://dataspheres.ai/mcp Claude Code can also install the server and the skills together as a plugin, which puts the skills in your project rather than behind the connection: /plugin marketplace add geekdreamzz/ari-dai-skills /plugin install dataspheres@ari-dai-skills Cursor Cursor supports a one-click install link. Add Dataspheres to Cursor opens Cursor with the server pre-filled; confirm it, then sign in when Cursor asks. The same configuration can be written by hand into .cursor/mcp.json in a project, or into the global file from Settings → MCP : { "mcpServers": { "dataspheres": { "url": "https://dataspheres.ai/mcp" } } } VS Code and GitHub Copilot VS Code has the same kind of link: Install Dataspheres in VS Code (or in VS Code Insiders ). VS Code asks you to trust the server and then runs the sign-in. Copilot's agent mode uses the tools from then on; the skills are in the Add Context menu as prompts. To configure it by file instead, create .vscode/mcp.json in the workspace: { "servers": { "dataspheres": { "type": "http", "url": "https://dataspheres.ai/mcp" } } } One limit worth knowing: VS Code allows at most 128 tools across all connected servers. Dataspheres stays well under that by listing the core set and reaching the rest through search_tools . OpenAI Codex The Codex command line adds the server with one line, or with two lines in its configuration file: codex mcp add dataspheres --url https://dataspheres.ai/mcp # ~/.codex/config.toml [mcp_servers.dataspheres] url = "https://dataspheres.ai/mcp" Codex handles the OAuth sign-in itself on recent versions; on older ones, an API key from the Developer Portal can be sent as a bearer token instead. Checking that it works The Connect tab of the Developer Portal has every link and command above, plus a connection test. The test runs the same sign-in an assistant runs and then asks the server for your dataspheres, so a successful list there means any assistant pointed at the address will see the same thing. In the assistant itself, the simplest check is to ask it to list your dataspheres; it will call get_context and answer with the names and your role in each. What the assistant can and cannot do The connection is scoped to your account, and your role is checked live in every datasphere on every call, so an assistant cannot do anything you could not do yourself. Actions that delete, send, publish or invite need an explicit confirmation that the assistant may only give after you have agreed. Review gates in content pipelines (approving a deck or a script before a video is made) stay with you: the assistant can show you what is waiting and record your decision, never make it. Disconnecting is d