Connect the agent you already use.
ChatGPT, Claude, Cursor, or an agent you run yourself. Add onti to it once, say yes in an onti window, and from then on it can look into an order when you ask.
Your agent never sees your password, and nothing reaches a customer until you say yes.
Three moves, whichever app you use.
Add onti to your app.
One address, pasted where your app keeps its connections. The exact clicks for ChatGPT, Claude, Cursor, and your own agent are below.
Sign in and say yes.
An onti window opens. Sign in or create your account, pick your business, and allow access. Keep that window open until it finishes. onti never asks for a password, a link, or a code in the chat.
Turn it on and check it.
Enable onti in a new conversation, then paste the check message. Your agent reports back with which app connected and when.
The address to add is the same everywhere:
https://api.onticord.ai/api/mcpConnect ChatGPT
Set up in ChatGPT web. Resume these instructions on your computer if needed.
Before you start
- Developer mode must be available for your account and workspace policy. Ask your workspace administrator if it is unavailable.
- ChatGPT needs a public HTTPS MCP endpoint or an explicitly configured Secure MCP Tunnel. Pasting localhost does not create a tunnel. onti does not provision one here.
In ChatGPT web, open Settings → Security and login and enable Developer mode.
Open Plugins and select the plus button. Name the connection onti. For a public connection, enter https://api.onticord.ai/api/mcp as the MCP server URL and use OAuth. If you already have a Secure MCP Tunnel configured, use its documented Tunnel connection instead.
https://api.onticord.ai/api/mcpWhen the onti window opens, sign in or create your account, pick your business, and allow access. Keep the window open until it finishes.
Start a new ChatGPT conversation. Open its tools menu and add the onti MCP connection before sending the connection-check prompt.
Paste the check message. Your agent confirms which app connected and when.
Connect Claude
Set up in Claude web. Resume these instructions on your computer if needed.
Before you start
- Individual Pro/Max accounts can add a custom connector. For Team/Enterprise, an Owner or Primary Owner must add it for the organization before members connect.
- Claude remote connectors run from Anthropic infrastructure, including desktop and mobile. The endpoint must be reachable from that infrastructure; localhost, VPN-only or private-network access from your own device is insufficient.
Individual account: open Customize → Connectors → + → Add custom connector. Name it onti and enter https://api.onticord.ai/api/mcp.
Team/Enterprise owner: open Organization settings → Connectors → Add → Custom → Web and add https://api.onticord.ai/api/mcp. Then each member opens Customize → Connectors, finds onti, and selects Connect to authenticate.
https://api.onticord.ai/api/mcpWhen the onti window opens, sign in or create your account, pick your business, and allow access. Keep the window open until it finishes.
In the Claude conversation, open + at the lower left, select Connectors, and turn on onti before sending the connection-check prompt.
Paste the check message. Your agent confirms which app connected and when.
Connect Cursor
Use Cursor on the computer where you want the connection. The manual path below configures your user account, not a cloud agent or every teammate.
Before you start
- Your Cursor configuration and team policy must allow this MCP server. A local agent can prepare the exact configuration change for you to review.
- Cursor supports Streamable HTTP with OAuth. A local endpoint requires Cursor to run on the same computer; web or cloud agents need their own reachable endpoint.
For an existing ~/.cursor/mcp.json, add only the server entry below inside mcpServers. Use the new-file example only when that file does not exist.
Open Customize in the Cursor sidebar to find and enable the server. Complete the OAuth flow Cursor presents for the server; if no sign-in prompt appears, use the official Cursor guide to check this server's setup. This guide provides manual configuration, not a marketplace installation.
https://api.onticord.ai/api/mcpAdd this entry inside your existing mcpServers object:
{ "onticord": { "url": "https://api.onticord.ai/api/mcp" } }If ~/.cursor/mcp.json does not exist yet
{ "mcpServers": { "onticord": { "url": "https://api.onticord.ai/api/mcp" } } }Keep every other server and setting in that file as it is. If a local agent edits it for you, it should show you the exact change first, add no tokens, and leave approvals on.
When the onti window opens, sign in or create your account, pick your business, and allow access. Keep the window open until it finishes.
Enable onti in Customize, then open Agent chat and check that its tools appear under Available Tools. Send the connection-check prompt there; keep required tool approvals enabled.
Paste the check message. Your agent confirms which app connected and when.
Connect your own agent
Use the settings of the client you manage, including self-managed Hermes or OpenClaw. Resume on the device running that client when needed.
Before you start
- The client must support Streamable HTTP and OAuth, and securely store its own credentials. A web reader cannot install tools just by opening this guide.
- The machine running your client must reach the MCP and authentication endpoints. Local-only clients may require a separately configured compatible bridge; none is installed by this guide.
Add a remote MCP server named onti using https://api.onticord.ai/api/mcp, Streamable HTTP transport and OAuth authentication. Follow your client's documentation to start its OAuth flow.
https://api.onticord.ai/api/mcpWhen the onti window opens, sign in or create your account, pick your business, and allow access. Keep the window open until it finishes.
Enable the onti server and its tools in the conversation using your client's documented controls. Check that the onti tools are available before sending the connection-check prompt.
Paste the check message. Your agent confirms which app connected and when.
Your agent needs remote MCP with OAuth and a safe place to keep its own credentials. Setup notes for OpenClaw and self-managed Hermes.
Check the connection.
With onti turned on in the conversation, paste this. Your agent does the checking and tells you which app connected and when. Reading your store on its own does not count.
Call get_agent_connection_setup and follow its nextAction. If needed, call start_agent_connection_setup, then explicitly call verify_agent_connection with setup.attempt.id from the successful result as setup_id. Reading business data alone does not complete this connection check. Report the actual OAuth client and dated evidence. If an explicit setup_id is expired or superseded, explain the result; never silently verify a replacement.
Then ask what it can see.
Once connected, this asks what is ready for looking into one late order and what still needs setting up. It changes nothing in your store.
Use onti's get_customer_rescue_readiness tool with no arguments to check my setup for one delayed-order investigation. Explain what is configured, what remains unverified, and the next action or operator handoff. Do not change setup or start an investigation from this check. Ask for an explicit investigation request and one order reference before starting a workflow.
Started on your phone?
Sign in on the computer that runs your app and open your saved setup. It picks up where you left off. The link carries your place, never your password.
Open your saved setupBuilding your own agent?
Everything an agent needs to read is written for it: how to request a sign-in link for its owner, how a custom client authorizes, and the check that proves a connection. Sign-in, choosing the business, and allowing access always stay with the person.