Docs · ChatGPT Connector

Connect Loopthink to ChatGPT.

One connector URL gives ChatGPT governed access to every MCP server your workspace has granted you — databases, HubSpot, Salesforce and curated tools, with field-level masking applied before any result reaches the model.

Setup — about two minutes

  1. 1

    In ChatGPT, open Settings → Connectors → Advanced → Developer mode, then Create a connector.

    Custom MCP connectors require a ChatGPT plan with developer mode (Plus/Pro/Business/Enterprise/Edu).

  2. 2

    Paste the connector URL https://mcp.eu.loopthink.ai/mcp and confirm.

  3. 3

    Sign in with your Loopthink account and, if you belong to several, pick a workspace.

    The connection is scoped to that workspace — reconnect to switch.

  4. 4

    Done. ChatGPT lists every tool your member group was granted — ask away.

Connector URL

https://mcp.eu.loopthink.ai/mcp

Example prompts

ChatGPT discovers your tools automatically. Each enabled table or CRM entity publishes its own tools — these examples exercise different ones:

  • “Which tables can you access, and what do they contain?”

    Uses the discovery tool (listTables / listEntities) — the fastest way to see what a workspace exposes.

  • “List all organisations with their member counts, sorted by creation date.”

    Combines read tools across tables — scan, then join by key.

  • “Look up the order with ID o-2024-1387 and summarize its status history.”

    Uses a get-by-primary-key tool with the table’s real key attributes.

  • “Search HubSpot contacts for “Miller” and show their companies.”

    Uses a CRM search tool — masking rules apply before ChatGPT sees the data.

Authentication

The connector uses OAuth 2.0 with PKCE and dynamic client registration. You sign in with your Loopthink account and pick a workspace; no API keys are shared with the AI client. Tokens are short-lived and refresh automatically.

Governance built in

ChatGPT only sees tools your member group was granted. Field-level masking rules are applied on the server before results reach the model, and every tool is annotated read-only unless writes were enabled explicitly.

Questions or issues? See our privacy policy or contact us via the details in the imprint.