Skip to content

MCP and API Keys

ShareCube speaks MCP. Agents can list projects, publish artifacts, search, and leave comments, then return a console URL so a human can keep editing.

Open Settings → Integrations to set this up.

Integrations and MCP setup

The default is Sign in. The first time a client connects, you complete a browser login and pick an organization. There is no secret in a config file, and you can revoke access later from the client or by signing out.

This is the right mode for Cursor, Claude Desktop, Copilot, and anything else where a person is sitting at the machine.

Use API key when nobody can finish a browser sign-in: CI, a container, a headless box.

  1. In Settings → Integrations, name the key (for example ci-runner) and click New key. Copy it immediately. ShareCube will not show the full value again.

  2. The snippets below the key list update to include Authorization: Bearer <your-key>.

  3. Use the trash icon on a key row. Admins can also review and revoke keys created by other members under organization keys.

ShareCube ships ready-made snippets for these clients. Expand a row to copy the config, or use Add to Cursor / Add to Copilot when the one-click install link is available.

ClientConfig file / command
CAPAcapabilities.yaml (set the key via capa install, never commit it)
Claude Codeclaude mcp add --transport http sharecube "https://app.sharecube.io/mcp"
Claude Desktopclaude_desktop_config.json
Codex.codex/config.toml
Copilot.vscode/mcp.json
Cursor.cursor/mcp.json
Gemini CLI.gemini/settings.json
Windsurf~/.codeium/windsurf/mcp_config.json

Cursor MCP snippet

The MCP endpoint is https://app.sharecube.io/mcp (use https://app-dev.sharecube.io/mcp against the dev environment).

{
"mcpServers": {
"sharecube": {
"url": "https://app.sharecube.io/mcp"
}
}
}

These are the MCP tools ShareCube exposes. The caller needs the same project role a human would need for the same action.

Projects and artifacts

  • list_projects / create_project
  • list_artifacts / get_artifact / create_artifact / update_artifact / delete_artifact
  • request_artifact_upload for large bodies (the tool mints an upload URL; the client PUTs the bytes)
  • search_artifact / search_graph

Comments

  • list_comments / get_comment
  • create_comment (new thread with an optional selection anchor, or a reply)
  • update_comment / resolve_comment / delete_comment

For large documents, prefer request_artifact_upload and then poll get_artifact until current_version_id changes and the content matches. For small edits and title changes, update_artifact is enough. Always pass base_version_id when updating content so a concurrent human edit is not overwritten.