Docs
Conducktor is the AI environment you own. Your connectors, the APIs you wire in, the tools you built with Claude, your files and the Assistants you created live in one workspace under your account. Any MCP client plugs into that workspace through one address. Switch apps or models and it is all still there.
https://mcp.conducktor.com/mcp
Paste this where your AI client asks for an MCP server URL, then sign in with your Conducktor account. No key to generate, nothing to host.
Claude on the web, Claude Desktop and Cowork share one list of connectors.
https://mcp.conducktor.com/mcp. Click Continue.On a Team plan only an Owner can add a connector, under Organization settings → Connectors; on Enterprise, an Owner or a role with Manage access to Libraries. Free accounts get one custom connector. The claude_desktop_config.json file is for local servers only; a remote server is added here, not there.
Add the server from the terminal, then sign in.
claude mcp add --transport http conducktor https://mcp.conducktor.com/mcp
Then run /mcp inside Claude Code and follow the sign-in link.
ChatGPT adds remote MCP servers in Developer mode.
https://mcp.conducktor.com/mcp, choose OAuth, and save.Cursor reads MCP servers from a JSON file.
File: .cursor/mcp.json (this project) or ~/.cursor/mcp.json (everywhere)
{
"mcpServers": {
"conducktor": {
"url": "https://mcp.conducktor.com/mcp"
}
}
}Cursor opens the sign-in page the first time it connects.
VS Code reads MCP servers from mcp.json.
File: .vscode/mcp.json
{
"servers": {
"conducktor": {
"type": "http",
"url": "https://mcp.conducktor.com/mcp"
}
}
}Or run MCP: Add Server from the Command Palette and paste the URL.
Any client that supports remote MCP servers over HTTPS with OAuth works the same way.
https://mcp.conducktor.com/mcp where the client asks for an MCP server URL.The official MCP Inspector opens a local page where you can call the server by hand.
npx @modelcontextprotocol/inspector
Then connect to https://mcp.conducktor.com/mcp.
Your client sees one tool, ask_conducktor. You say what you want in plain words; Conducktor runs it with the connectors, tools and Assistants in your workspace and returns the result. Reads run straight away. Whether a write needs your confirmation is the Assistant's autonomy level in Conducktor: Auto asks nothing; Supervised asks before a delete, a send outside your workspace or a spend. A confirmation cannot be given from an external client: Conducktor tells your client it has to come from you (a no from your client cancels the action), and nothing runs. Do the action in Conducktor itself. Today a Supervised Assistant holds every write it is asked for from an external client.
OAuth authorization code with PKCE (S256 recommended; required for public clients). Your client registers itself with Conducktor; no shared secret is pasted anywhere.
You approve the connection on a Conducktor page, signed in as yourself. The token your client receives is tied to your account.
Conducktor issues the scopes mcp:tools, mcp:read and mcp:write. You can see every connected client, and disconnect one, on your Account page.
Conducktor reaches the services you connected, the files you put in your workspace, and public web pages when you ask it to. Conducktor calls connected services with credentials held in its vault; your AI client receives the result of what you asked. Nothing is shared with other Conducktor users unless you bridge a connection to another workspace or share a link.
https://mcp.conducktor.com/mcp, HTTPS only.Email support@conducktor.com or visit .
Signed in already? Your URL and these steps are also on your Account page under AI Clients.