Add Signl in ChatGPT's connector settings using the MCP server URL from your dashboard, then approve access when prompted. Setup takes about two minutes. This flow is currently marked beta: it follows OpenAI's documented connector path, but we have not verified it end to end, so use the API key fallback if anything does not match.
Before you start
You need an accepted application and a Signl account, and a ChatGPT plan that includes custom connectors. Rather than checking a plan list that changes, open Settings and look for Connectors. If you can add a custom one, you are set.
Have your MCP server URL ready. It is on the Connect page of your dashboard.
Steps
- In ChatGPT, go to Settings, then Connectors.
- Click Create, or Add connector.
- Name it Signl and paste the MCP server URL from your dashboard.
- Follow the sign-in prompt and approve access when asked.
- Enable the Signl connector inside a chat, then ask it to find leads.
Step 5 is the one people miss. Adding a connector in settings does not always switch it on for conversations that were already open. Start a new chat and confirm Signl is enabled for it.
Confirm it worked
Ask:
What Signl tools do you have access to?
It should list five: search_signals, get_weekly_drop, get_credit_balance, enrich_lead and refresh_person_signals. If it lists none, see my agent cannot see Signl's tools.
Then try a real one:
Show me this week's hottest signals.
That calls get_weekly_drop, which is free on every plan, so it is a safe first test.
If the interface does not match
OpenAI moves these labels, and connector availability has changed by plan more than once. If the steps above do not match what you see, do not force it. Use an API key instead, which works regardless of plan and does not depend on the connector UI:
- In your dashboard, open API Key and generate one. It begins with
sk_signl_. - Copy it immediately. It is shown exactly once and stored only as a SHA-256 hash.
- Add it as an
Authorization: Bearerheader in whichever client you are configuring.
See connect Cursor and the CLI for the config shape, which is the same anywhere a bearer token is accepted.
Known differences from Claude
Worth knowing if you use both.
- Claude's flow is verified; this one is not. If you want the path we are confident in, it is connect your agent.
- Per-chat enabling. Claude applies a connector once authorised. ChatGPT wants it switched on in the conversation.
- Undocumented limits. Anthropic publishes its maximum tool result size and timeout. OpenAI's MCP documentation does not, so on ChatGPT you find the ceiling by hitting it. Ask for twenty results rather than a hundred until you know where it sits, and ask how many rows came back before asking for analysis.
Related
Connect your agent for the overview and the OAuth path, the MCP tool reference for what each tool takes, and my agent cannot see Signl's tools if nothing appears.