ChatGPT MCP and Claude web connectors with rConfig: what works today

ChatGPT can’t connect to rConfig’s MCP server. Claude on the web can, with conditions. The difference is authentication: rConfig accepts a bearer token, ChatGPT can’t send one, and claude.ai can (in a beta). Both run in their vendor’s cloud, so either way rConfig has to be reachable from the internet.

Why the two differ

	ChatGPT	claude.ai (and Claude Desktop custom connectors)
Connects from	OpenAI’s cloud	Anthropic’s cloud
Authentication options	OAuth, no authentication, or a mix of the two	OAuth, no sign-in, and fixed Request headers (beta)
Can send rConfig’s bearer token	No. OpenAI states ChatGPT can’t present custom API keys	Yes, as an `Authorization: Bearer` request header
Works with rConfig	No	Yes, if `/mcp` is internet-reachable and your organisation has Request headers

Checked against OpenAI’s and Anthropic’s documentation on 7 November 2026. We’ll update this page if either changes.

What to use instead

On OpenAI models: use Codex CLI. It runs on your machine or server, sends the token from an environment variable, and reaches an internal rConfig directly.

On Claude, with an internal rConfig: use Claude Desktop with the local bridge or Claude Code. Both connect from inside your network.

Setting up a claude.ai custom connector

If rConfig’s MCP endpoint is already published, or you decide to publish it:

Publish /mcp only. Put rConfig behind a reverse proxy that exposes the /mcp path and nothing else, over HTTPS with a public certificate. Restrict source addresses at the proxy or firewall where you can.
Create a dedicated token. Make an rConfig user whose role covers only the devices the connector’s users should see, and create an MCP token owned by that user: an admin picks the user under Token for on the MCP Server tab. Choose an expiry, because a request header is one credential shared by everyone who uses the connector.
Add the connector. In Claude, open Customize → Connectors (on Team and Enterprise plans, an Owner uses Organization settings → Connectors), add a custom connector with the /mcp URL, choose No sign-in, and add an authorization request header with the value Bearer <token>.

The full settings are on the Claude Desktop page.

Frequently asked questions

Can I connect ChatGPT to rConfig?

Not today. ChatGPT's custom MCP connections support OAuth or no authentication, and OpenAI states that ChatGPT can't present custom API keys. rConfig's MCP server requires a bearer token. Use Codex CLI to work with rConfig on OpenAI models.

Can claude.ai connect to rConfig?

Yes, if rConfig's MCP endpoint is reachable from the internet and your organisation has the Request headers beta for custom connectors. The connector sends the token as an Authorization: Bearer header from Anthropic's cloud.

What should I use if rConfig is on an internal network?

An agent that runs inside your network: Claude Code, Claude Desktop with the local mcp-remote bridge, Cursor, VS Code with Copilot, Codex CLI or a self-hosted n8n.

Can I set rConfig's MCP server to no authentication for ChatGPT?

No. rConfig always requires a token on /mcp, and an unauthenticated endpoint returning device configurations would be a serious exposure.