Connect · ChatGPT and OpenAI

Use SignalPilot with ChatGPT and OpenAI

SignalPilot speaks the same MCP standard OpenAI uses, so ChatGPT, Codex, and agents built on the OpenAI SDK can all talk to your data through the same governed endpoint.

All three paths use the same endpoint and the same API key. Create the key once under Settings → API Keys in SignalPilot; it starts with sp_ and is shown only when created.

Endpoint

https://gateway.signalpilot.ai/mcp

Header

x-api-key: sp_your_key_here

Bearer tokens work too: Authorization: Bearer sp_…

ChatGPT

ChatGPT adds MCP servers as connectors in developer mode.

1. 1

   Turn on developer mode

   In ChatGPT, open Settings → Connectors → Advanced settings and switch on Developer mode. A workspace admin may need to enable this for you.

2. 2

   Create the connector

   Back on Connectors, choose Create. Enter SignalPilot as the name, paste the endpoint above as the MCP server URL, and confirm.

3. 3

   Sign in when asked

   ChatGPT connectors cannot send a custom API key header. Instead, choose the sign-in option when ChatGPT offers it and log in with your SignalPilot account. The connector then acts as you, with your workspace's permissions.

4. 4

   Use it in a chat

   Start a new chat, open the tools menu, and select SignalPilot. Ask a question you can verify, such as how many orders shipped last week.

Sign-in availability

If your ChatGPT workspace only offers "No authentication" for custom connectors, contact your SignalPilot admin: the workspace needs sign-in enabled on the gateway before ChatGPT can connect. Codex and the Agents SDK below work with a plain API key today.

Codex

Codex has a SignalPilot plugin with the same skills and verifier agents as the Claude Code plugin. Install it, then point it at the cloud endpoint.

install the codex plugin

$ codex plugin marketplace add SignalPilot-Labs/codex-signalpilot-plugin
$ codex plugin add signalpilot@signalpilot

Add the server to ~/.codex/config.toml:

~/.codex/config.toml

[mcp_servers.signalpilot]
url = "https://gateway.signalpilot.ai/mcp"
http_headers = { "x-api-key" = "sp_your_key_here" }

Restart Codex. Ask it to list the SignalPilot tools to confirm the connection. The full plugin guide is on the Codex page.

OpenAI Agents SDK

Register SignalPilot as a hosted MCP tool. The OpenAI platform calls the endpoint for you, so no local process is needed.

python

from agents import Agent, HostedMCPTool
agent = Agent(
    name="analyst",
    instructions="Answer questions using SignalPilot. Show the SQL you ran.",
    tools=[
        HostedMCPTool(
            tool_config={
                "type": "mcp",
                "server_label": "signalpilot",
                "server_url": "https://gateway.signalpilot.ai/mcp",
                "headers": {"x-api-key": "sp_your_key_here"},
                "require_approval": "never",
            }
        )
    ],
)

Keep the key in an environment variable rather than in source. Governance runs on SignalPilot's side, so the agent can only read what the workspace allows regardless of what the prompt asks for.

Next

• Connect a database if the workspace has none yet.
• API keys and permissions for scopes, rotation, and rate limits.
• Claude and Claude Code setup, if your team uses both.