tiqline is a process SLA monitoring tool: it turns events from Shopify and your own systems into one case per piece of work, gives every step a time limit and alerts you in Slack or by email when a step runs over.
Set up process monitoring with an AI agent: describe the process, the agent builds it
Most people know their process better than any configuration screen. With tiqline's MCP server you describe it in plain words to an AI agent such as Claude, ChatGPT or Grok. The agent looks at your real events, proposes the steps, and after your go-ahead creates the process, checks how your real cases run through it, sets the limits and connects the alerts.
Why setting up monitoring takes longer than it should
The hard part of process monitoring is rarely the alert. It is the model: which event opens a case, which property identifies it, which steps follow in which order, where the paths split and which limit applies to each step. Getting that right means looking at real events, trying a definition, seeing which cases end up where and adjusting it. In a classic tool that is an afternoon of forms. In a self-built automation in Zapier or n8n it is a workflow per rule plus the state you keep yourself.
What MCP changes
The Model Context Protocol is an open standard that lets AI applications use tools from other services. Claude, ChatGPT, VS Code and Cursor support it. With an MCP server, the agent does not just write you instructions; it can read your data and make the change, with you deciding what it may do.
What the agent needs before it starts
The agent works with what tiqline already receives. Connect your Shopify shop in the dashboard and let your other systems send their events through the track API first. Then the agent can see the real event names and properties instead of guessing them. With set_id_fields the agent says which properties hold the IDs of which objects, for example that order_id holds an order and shipment_id a shipment. An event that carries both IDs links the order and its shipment, and those links are what tiqline builds cases from. The agent should ask you before it changes this, since it replaces the whole mapping and rebuilds all cases.
How an AI agent sets up a process in tiqline
1. Connect the agent
Claude, ChatGPT and Grok connect to tiqline's MCP server through OAuth, without an API key. Admins see which apps are connected and can disconnect each one.
2. The agent reads your events
Eight tools only read: list_event_types, list_property_values, preview_journeys, list_processes, get_recompute_status, get_audit_log, get_usage and list_incidents. With them the agent sees which events arrive and which values a property takes before it proposes anything, and later how cases run through a process and which incidents are open.
3. The agent proposes and you confirm
Six tools write: set_id_fields, create_process, set_sla, set_alert_channel, archive_process and unarchive_process. They are marked as writing tools, so apps that support it ask you before the agent uses them. A good habit is to let the agent describe the proposed steps, branches and limits in plain words first, and to give the go-ahead only then.
4. The process is built from stored events
When the agent creates the process, tiqline builds the cases from the events it has already stored. For an organisation with up to 5,000 events that is done when the call returns; larger ones are built in the background, and the agent can follow the progress. A definition that is not valid is refused with an error message, so the agent can correct it. Then the agent calls preview_journeys: it shows how many cases are running, stuck or done, a sample of them, and warnings about the data, such as objects that belong to no case.
5. Limits, alerts and later changes
The agent sets limits per step with set_sla and connects Slack, email or a webhook with set_alert_channel. Later you can ask it to change a limit, and the change applies at once to cases already waiting. Every plan includes the audit log, including Free. Publishing a process, changing a time limit and connecting Slack or email are written to it, with the person or app that did it.
6. Agents that react to incidents
An agent can read open incidents with list_incidents. Or you point the webhook channel at an automation, such as a Cursor Automation's webhook trigger, so an agent starts work as soon as an incident opens.
Example: a conversation that ends with a working process
The shop, the events and the limits are made up to show the flow.
You: watch our orders from payment to shipping
You tell the agent: orders should be packed within 4 hours after payment in Lager Nord and within 8 hours in Lager West, and shipped within 24 hours after packing. Alerts go to Slack.
The agent looks at the data
It lists the event types and finds Order Paid from Shopify and Order Packed from the warehouse system, with a property warehouse that takes the values north and west. It proposes a process with a branch on that property, and you agree.
The agent builds and checks
The agent creates the process and calls preview_journeys. The cases are built from the orders already stored. Orders without a warehouse value take the else path, which the agent gives 8 hours with set_sla after asking you.
Alerts are connected
Stuck
The agent connects the Slack channel with a test message. Three orders from this morning are already past their packing limit and show up as stuck.
What the agent cannot do
The MCP server has no tool to connect a Shopify shop, invite people or change the plan; those happen in the dashboard. The agent cannot delete a process, only archive it and bring it back. It can only work with events that tiqline already receives, so a step no system sends stays invisible to the agent too. Whether an app asks you before it runs a writing tool depends on the app; tiqline marks the writing tools. Every plan includes the audit log, including Free. Publishing a process, changing a time limit and connecting Slack or email are written to it. The agent proposes limits, but which limit is right for your business is your decision.
When to use the dashboard instead
For a small change, such as one limit, the dashboard editor is just as quick: draft, preview on real cases, publish. The agent pays off when you start from nothing, when the events are new to you, or when you want to change several things in one go.
tiqline monitors process SLAs and is not one of the classic product-analytics tools.
Questions about setting up tiqline with an AI agent
Which AI apps can connect?
Claude, ChatGPT and Grok connect through OAuth without an API key. Other clients that support MCP can use the server as well.
Can the agent change things without asking me?
The six writing tools are marked as such, so apps that support this ask you first. Every plan includes the audit log, including Free. Publishing a process, changing a time limit and connecting Slack or email are written to it, with the person or app that did it.
Which tools does the MCP server have?
Fourteen: eight that read, such as list_event_types and preview_journeys, and six that write, such as create_process, set_sla and set_alert_channel.
Can I see what the agent changed?
Yes. Every plan includes the audit log, including Free. It records publishing a process, changing a time limit and connecting Slack or email, with who did it, and get_audit_log lets an agent read it.
Can I disconnect an agent?
Yes. Admins see the connected apps and can disconnect each one.
What does set_id_fields do?
It says which event properties hold the IDs of objects, such as order_id for an order. Events that carry the same ID, or two IDs at once, are linked, and cases are built from those links.
Does the agent need my Shopify access?
No. You connect the shop in tiqline's dashboard. The agent only works with the events tiqline has stored.
Start for free
The Free plan includes 5,000 events a month.