What you'll do
By the end of this guide, ChatGPT reads your SignalRaven workspace and answers questions about it: who showed intent this week, why a signal matters, and what to say. You add one server, approve one connection, and send one prompt.
Before you start
- The ChatGPT desktop appfor Mac or Windows. The menus below are the desktop app's.
- A SignalRaven account. A trial or paid workspace gives ChatGPT live signals. Any other account gets a fully labeled sample workspace, a good way to try the connection first.
- If you're already signed in to signalraven.ai in your browser, the sign-in step skips straight to the consent screen.
Setup
Open the MCPs tab
In ChatGPT, open Plugins in the sidebar, then the MCPs tab. Every custom server you add lives here, each with a gear icon for its settings.
Add the server
Press Add server. Name it SignalRaven, choose Streamable HTTP as the transport, and paste the URL. Leave the token and header fields empty; sign-in happens in the next step. Save.
Authenticate
Press Authenticate on the new entry. SignalRaven's sign-in opens in your browser. Sign in with Google, Microsoft, or a work-email link, then press Allow on the consent screen. Back in ChatGPT, the entry shows as connected.
The consent screen lists what ChatGPT gets: reading your signals, sources, watchlist, ICP, intelligence reports, usage and destinations, plus starting report runs and posting to destinations. ChatGPT asks you before it starts a run, since runs spend credits.
Connect ChatGPT
Sign in and you head straight back to finish connecting.
Switch to Work
Custom servers attach to Work sessions. At the top of the window, switch from Chat to Work and start a new session.

Turn SignalRaven on for the session
Open the Plugins chip under the composer and switch SignalRaven on. Pick any model except Instant. Instant answers from its own knowledge and skips tools.
Send your first prompt
ChatGPT calls the server, reads your signals, and answers with a ranked list: the person, their title and company, a strength score out of 10, why it matters, and an opener for each.

What to ask next
Name the tool in the prompt and ChatGPT reaches for it every time. These five cover most of a working day:
- Show my strongest signals this week and draft an opener for each.
- Which named accounts showed activity this week, and through which source?
- Research [person] before my call: what did they engage with, and what should I say?
- Which of my monitored sources are producing qualified signals, and which are dead?
- Run Account Intelligence on [company] and tell me who to email first.
The Skills page covers the two packaged workflows, research a prospect and draft openers, and the phrases that trigger them.
If something's off
“SignalRaven isn't exposing its signal feed to me in this chat right now.”
The server isn't attached to this session. Start a new session in Work mode and turn SignalRaven on in the Plugins chip. Chat mode leaves custom servers off.
“The connector is installed, but its token is missing the read:signals permission.”
The connection was approved before permissions were requested. Delete the server, add it again under a new name, and authenticate. Re-adding under the same name reuses the old connection and skips the consent screen.
ChatGPT answers from its own knowledge instead of calling SignalRaven.
Switch the model away from Instant, and start the prompt with “Using SignalRaven”.
I see sample data instead of my workspace.
The account you signed in with has no active SignalRaven workspace, or you approved the connection as a different user. Delete the server, sign out of signalraven.ai in your browser, and connect again as the account that owns the workspace.
Every message users have hit so far, quoted verbatim, is on the ChatGPT connector page.
Where to go next
- ChatGPT + SignalRaven → The connector page: prompts, skills, troubleshooting, and the demo.
- Skills → Research a prospect and draft openers, step by step.
- Codex → The same server from your terminal or editor.
- MCP reference → OAuth, scopes, tools and output schemas for any client.