Table of contents
Below are 11 MCP use cases that go past „find this conversation.“ Some are analytics, some clean up the mess in your tags, some just save you twenty minutes of clicking.
What Smartsupp MCP can access
Right now, the Smartsupp MCP server exposes 18 tools. That number keeps growing, so treat it as a snapshot rather than a limit. New tools get added as the product develops, and your assistant picks up whatever is available when it connects.
What matters more than the count is the shape of it.
| Area | What you can do today |
|---|---|
| Statistics | Query 7 metrics, grouped by any of 17 dimensions |
| Conversations | Search, filter, read messages, reply, close, reopen, email a transcript |
| Contacts | Find a contact by email, pull their full profile |
| Management | List and create tags, tag and untag conversations, see operator availability, assign and unassign |
The metrics: response time, first response time, time to close, new conversations, closed conversations, conversation ratings. The dimensions you can group them by include day, week, month, day of week, hour of day, operator, tag group, rating value, and whether a conversation happened outside your business hours.
You never type any of these names. Your assistant translates a plain-language question into the right calls. They are listed so you can see the range.
One question, several queries
Statistics take one grouping dimension per query. So „which weekday has my worst first response time, and what was going on that day“ is not a single query. It is two or three, and your assistant runs them in sequence and assembles the answer. In a dashboard you would export first response time by weekday, export conversation volume by weekday, then line them up in a spreadsheet.
11 MCP use cases that replace an export and a spreadsheet
Chat data tells you more about your business than most people get out of it. Every conversation carries the page the customer was on and what they had browsed before. It also records how long they waited, who answered, and whether they were happy afterwards. That is a record of where your website confuses people and where your team runs out of capacity, and it is sitting in your account right now.
What follows are eleven questions worth asking it, grouped by what they get you. Three find money you are currently leaving on the table. Four are about your team and where it runs out of capacity. Two clean up the tags your reporting depends on, because per-tag numbers are only as good as the tagging behind them. The last two are routines that take a sentence instead of five minutes of clicking.
If you are building out how your team measures itself, our guide to customer success metrics is a good companion to this list, and customer care teams covers how to organise around what you find.
One practical note before you start. Ask about a quarter or a year, not a week. Over one week, one person being ill looks like a pattern. Over a quarter it either is one or it isn’t.
Money you are leaving on the table
1. Find out which pages your questions come from
Go through conversations from the last quarter and tell me which pages and traffic sources customers were on when they started a chat. Which pages generate the most questions?
How it works
Pulls conversations for the period and reads the page and referrer recorded on each one, then ranks them.
What comes back
A ranked list of pages and referrers by conversation volume, and where those conversations take longest to resolve.
If a third of your questions come from one product page, that page has a description problem, not a support problem. Fixing the page removes the conversations instead of answering them faster. Our analysis of 4.78 billion website visits found that only 0.84% of visitors start a chat at all, so the ones who do are worth understanding.
2. Count the messages that arrive outside business hours, and how many got answered
How many conversations came in outside business hours last quarter, and how many of those got a reply at all?
How it works
Conversations grouped by whether they arrived outside business hours, then by whether an out-of-hours reply happened.
What comes back
The volume of out-of-hours conversations, and how many were answered.
The gap between those two numbers is the interesting one. It is the size of the problem in conversations, which you can weigh against the cost of an evening shift or an AI agent answering while nobody is there. In our 2024 data, stores running a chatbot handled 89.2% of inquiries against 71.2% without one.
One condition before you trust the answer. Out-of-hours is defined by the business hours set in your Smartsupp account, not by what your team actually does. Say your account says Monday to Friday and someone answers on Saturdays anyway. Those Saturday conversations count as out-of-hours, and the number will mislead you. Check the setting first.
3. Find conversations nobody ever replied to
Find conversations from the last quarter where the customer wrote and no operator ever replied. How many are there, and what were they about?
How it works
This one has no direct filter, so your assistant works through the conversations for the period and checks the messages on each. On a large account, narrow the date range or ask about one month at a time.
What comes back
The count, and the conversations themselves so you can see what you missed.
These conversations look closed, and closed looks fine, which is why nobody builds a report for them. For a store this is lost demand you can put a number on, and it is a different problem from replying slowly. Slow is a capacity issue. Silent is a coverage hole.
Your team and where it runs out of capacity
4. Find your slowest weekday, and what was different about it
Which day of the week has the longest first response time over the last quarter? Then tell me what else was different about that day. How many conversations came in, and how many operators were handling them?
How it works
First response time grouped by day of week, then two follow-up queries on the same period. Conversation volume by day, and operator activity by day. Your assistant lines them up.
What comes back
The weekday with the slowest first response, plus the context. Volume for that day, how many operators were active, how it compares to the rest of the week.
The number on its own points at whoever was on shift. The context points at the shift itself, which is the more useful conclusion. Be deliberate here: the output you want is a staffing decision, not a name. If one person does come back slower than everyone else, that is usually a workload or training question. Running the same query over a longer period will tell you which.
Weekends are included, which is either what you want or something to strip out depending on whether anyone works them. There is a real pattern to find: when we analysed all Smartsupp accounts, 80% of Monday inquiries got answered against 53.9% on Sundays.
5. Check whether your quality holds up in your busiest hour
Show me new conversations by hour of day for the last three months. For the busiest hours, also give me the first response time and the ratings.
How it works
New conversations grouped by hour of day, then first response time and ratings for the same window, compared against the peak hours.
What comes back
Your intake curve across the day, with response time and satisfaction in the peak hours.
Knowing when you are busy is mildly interesting on its own. Whether quality holds when you are busy is the question worth answering. If response times climb and ratings drop in the same hour, you have found a window where adding capacity pays for itself. If quality holds through the peak, your staffing is fine and the volume is just volume. If it doesn’t, handling several chats at once is the skill that gets tested in exactly that hour.
6. Compare how long each type of question takes to close
For each tag, what’s the average time to close? Which tags take the longest?
How it works
Pulls conversations per tag and averages time to close across each group. Statistics can’t be grouped by tag directly, so this is assembled tag by tag.
What comes back
Your tags ranked by how long those conversations stay open.
Average handling time is usually reported per operator, which turns a process question into a performance question. Ranked by tag, the same data says something more useful: this category of problem is structurally slow. That leads to a help article, a shortcut or a suggested reply from the AI Reply Assistant, none of which involve asking anyone to work faster.
7. Read your worst-rated conversations and find the pattern
Find the conversations with the worst ratings over the last quarter, read them, and tell me what patterns you see. Where do these conversations go wrong, and what should we change?
How it works
Finds low-rated conversations, reads the messages in each, and looks for what they have in common.
What comes back
The common failure patterns, the specific point where each conversation went off track, and suggested changes.
A rating tells you something dropped. It cannot tell you that four of the last ten bad ratings came after a customer was asked to repeat information they had already given. Reading twenty conversations to find that out is a job nobody has time for. It is also the only route from the number to the cause.
The tags your reporting depends on
8. Clean up the tags your reporting runs on
Go through my tags and the conversations from the last quarter. Which tags are duplicates or dead, which conversations have no tag at all, and what would you propose for each? Read the untagged ones and suggest a tag from my existing set.
How it works
Lists your tags, counts conversations per tag to see what is actually in use, then finds the conversations with no tag and reads them so it can propose one.
What comes back
Every tag with its usage count, sorted into duplicates, dead tags and near-dead ones, plus the untagged conversations with a proposed tag each. Ready to apply once you approve it.
The same question covers the whole cleanup, so you can point it at whichever part is your problem: merging duplicates (Sales, sales and Sales-inquiry are three entries that should be one), retiring dead tags nobody has used in a year, clearing an untagged backlog you were never going to work through by hand, or retagging the conversations affected by a merge you approved.
Worth running before questions 6 and 9, because both get more accurate afterwards. Untagged conversations are why your per-tag numbers are off, and a merge you keep postponing is why two of your tags look half as busy as they are.
9. Find recurring topics you have no tag for
Read the conversations from the last two months, work out what they’re actually about, and compare that against my existing tags. What themes come up repeatedly that I have no tag for? And are any conversations tagged wrongly?
How it works
Reads the message content of conversations for the period, clusters them by what they are about, and compares that against your tag list.
What comes back
Recurring themes with no tag of their own, plus conversations sitting under a tag that doesn’t match their content.
Reporting can only count categories you already invented, so a problem with no tag is invisible in every report you have. This question finds it because it works from what conversations say rather than from how they were filed.
The second half matters as much as the first. Tag statistics can be wrong even when everything is tagged, because the same issue ends up under two labels. Catching that needs someone to read the content, which is why it never gets done. Reading transcripts properly has always been the way to find this. The difference is that it no longer costs you an afternoon.
Routines that used to take clicking
10. See what the customer browsed before starting the chat
Find the contact with this email, pull their conversation history, and include the pages they visited before starting the chat.
How it works
Looks the contact up by email, retrieves their conversations, and includes the visitor browsing path on each one.
What comes back
The contact’s details, previous conversations, and the pages they moved through before starting the chat.
The browsing path is there but nobody thinks to look, and it is often the whole answer. Think of a customer stuck between two product variants, or one who read the returns policy three times before writing. They have told you what they want before typing a word. Useful before a call, and useful for a reply that lands first time.
One limitation: contact search works by email only, not by name or phone number. If you don’t have the email, find the conversation first and work back from it.
11. Reassign conversations from operators who are offline
Who’s online right now? Show me unresolved conversations assigned to operators who are offline, and reassign them to whoever’s available.
How it works
Checks operator availability, finds unresolved conversations assigned to people who are offline, and reassigns them.
What comes back
Current availability, the stuck conversations, and the reassignment.
Your dashboard will do this too, but it takes clicking. This is not a clever analytical trick. It is a routine you run at a shift change or when someone is unexpectedly out, and it takes one sentence instead of five minutes.
Good to know before you start
- It works with the data your account already has. It sees exactly what your account sees, no more.
- One grouping dimension per query. Crossing several means several queries, so a complex question takes longer.
- Statistics can’t be grouped by tag directly. Questions 6, 8 and 9 work by pulling conversations per tag and aggregating. Fine on most accounts, slower on large ones.
- Contact search works by email only. Not by name, not by phone number.
- Some questions have no direct filter and are answered by reading conversations one by one. Question 3 is the clearest example. Narrow the date range if it drags.
- Statistics reach back one year on Solo and Plus, three years on Expert.
- It works alongside your dashboard. For a number you check every week, use the dashboard. MCP is for the question you ask once.
How to connect Smartsupp MCP?
Smartsupp MCP is available on the Plus, Expert and Ultimate plans at no extra cost. Free and Solo don’t include the API access it needs.
It works with Claude, ChatGPT and Gemini, and each has its own setup guide in the Smartsupp MCP documentation. One thing to check first: your assistant needs its own paid plan too. Claude requires Pro or Max, ChatGPT requires Plus or above with Developer Mode enabled, and Gemini connects through the Gemini CLI.
Setup is the same three steps everywhere.
Add the Smartsupp MCP server to your assistant. In Claude Code that is one command:
claude mcp add --transport http smartsupp https://mcp.smartsupp.com/mcpAuthorise it. Your assistant opens a Smartsupp login page and you sign in as usual. Your password is never shared with the assistant.
Test it. Ask something small, like “list all agents on my Smartsupp account.”
Then start with any of the 11 MCP use cases above. There is more on what MCP does on the Smartsupp MCP page.
FAQs about Smartsupp MCP
Do I need to be a developer to use Smartsupp MCP?
No. Setup is one command or a few clicks depending on your assistant, and after that everything happens in plain language. You never write a query or learn a tool name. The prompts in this article are the whole interface.
Which plans include Smartsupp MCP?
Plus, Expert and Ultimate, at no extra cost. MCP runs on API access, which Free and Solo don’t include. Your AI assistant needs a paid plan of its own as well.
Can MCP change my data, or only read it?
Both. It can send messages, add and remove tags, assign operators and close conversations. Your assistant asks before it acts, so nothing changes without your say-so. If you only want to look, just ask questions.
Which AI assistants does this work with?
Claude, ChatGPT and Gemini, each with its own guide in the documentation. Any assistant supporting the MCP standard can connect.
How far back can I query statistics?
One year on Solo and Plus, three years on Expert. Worth knowing when you ask for year-on-year comparisons.