How do you use Riley with ChatGPT, Claude, Slack, and MCP?

Riley meets you where you already work. It connects to Claude Desktop, Claude Code, and Cursor as a tool you can call from inside them, works in ChatGPT where your ChatGPT plan supports custom connectors, and plugs into any other client that speaks the open connection standard called MCP. Slack is a delivery channel: receipts, approval requests, and alerts land there.

Wherever you ask, it is the same worker with the same rules. Anything that writes outside your workspace still waits for your approval, and the work still shows up in your receipts.

Slack is a first-party connection, so there is no key or endpoint to manage. Install the Riley app from your dashboard, invite Riley into the channels your team reads, and choose what gets delivered there. When Riley posts a draft or an approval request, you approve it right from the notification.

Drafts, approvals, and receipts land in Slack, so the people who need to see the work already see it.

These three are verified surfaces. You mint a Riley key in Settings, which is shown once, then add Riley as a connection in each tool. In Claude Desktop you paste the server details into the developer config. In Claude Code you add Riley with a single command. In Cursor you add it as a server in settings.

Once connected, you can ask any of them to use Riley, and the work appears back in your dashboard alongside everything else. External writes still queue for your approval, no matter which tool asked.

ChatGPT can use Riley where your ChatGPT plan supports adding a custom MCP connector. You mint a key, add a custom server in ChatGPT's connector settings, and paste the endpoint and key. Availability depends on your ChatGPT plan, so this one is honestly labeled as working where supported rather than everywhere.

The rules do not change. ChatGPT can call Riley's tools with your key, and anything external still waits for your approval.

MCP is an open standard that lets AI tools talk to services in a common language. Riley publishes a documented MCP endpoint with per-workspace keys, so any client that speaks MCP can connect: point it at the Riley endpoint and send your key as a header.

You can also bring your own tools. Register your own MCP server or import an OpenAPI spec in your dashboard, and Riley uses those tools under the same approval and receipt rules as everything built in.