Automation · Customer service and scheduling
Answer on WhatsApp with your Crisp knowledge base and what the customer already said in the chat
An AI agent answers on WhatsApp with your Crisp knowledge base articles and reviews what the customer already discussed in your website chat.
How do I make my WhatsApp answer with the Crisp knowledge base and know what the customer already asked in the website chat?
A Met connected to Crisp finds the knowledge base article that matches the question, reads it in full and answers with what it says. If the person had already written through your website chat, it finds them by name or email, lists their conversations and reads the recent messages so it does not ask the same thing again. If the case is still open in Crisp, it leaves an internal note with what was said on WhatsApp so your team sees it there, and updates the contact's segments based on what they asked for.
What starts it
When the customer writes
It starts when a customer writes on WhatsApp with a question about your service or about a case they already raised through your website chat. There is no trigger to configure: the Met assigned to the line runs the procedure inside that conversation.
Nothing to set up. The Met assigned to that channel picks up the conversation and runs the procedure inside it.
Who runs it
Support Met with a knowledge base in Crisp
Customer service team
This is not a named, seeded Met. It is the Met that already answers the business's WhatsApp, with Crisp connected. Its work ends when the customer gets the answer from your knowledge base and your team sees in Crisp what happened on WhatsApp.
Run by the customer support Met
Channels: WhatsApp
What you need connected
- Crisp Sell and serve
5 steps · 7 tools
The procedure, step by step
Each step shows the tool that runs. They belong to the real connector: if one of them stopped existing, this page would not build.
- 1
Find the article that answers
The Met searches the Crisp knowledge base for the file whose title matches the question, for example "How to change my plan". Search is by title, so your articles should have clear titles.
- 2
Read it in full and answer
The Met gets the article's content and answers with what it says, without inventing terms. If the article changes in Crisp, the next answer already uses the new text.
- 3
Find the contact in Crisp
If the person says they already wrote through the website chat, the Met looks them up in the Crisp CRM by full name or email and lists the conversations linked to them.
- 4
Read what they already said
The Met reads the recent messages of the website conversation and picks up from there. The customer does not have to repeat the order number or the problem they already explained.
- 5
Leave context for the team in Crisp
If the case is still open in Crisp, the Met leaves an internal note with what was said on WhatsApp and updates the contact's segments based on what they asked for. Only your team sees the note; the customer does not get it.
Before you start
What you need ready
The Met solves none of these for you. If one is missing, the automation stops there.
-
A knowledge base in Crisp
Your team has the knowledge base published in Crisp with clearly titled articles. Whatever is not written there, the Met does not know.
-
Crisp connected with its token
In Crisp, generate the MCP server token in Workspace Settings → Advanced configuration. In Met, open Connections → Crisp → Connect, paste the token and leave notes and segments switched on.
-
A WhatsApp line with the Met assigned
The business connects its WhatsApp Business API line and assigns the Met that answers customers. The answer happens in that conversation; Crisp remains the source of knowledge and history.
The limits
What this automation does NOT solve
Better said here than discovered halfway through the rollout.
-
This procedure does not write or fix articles. If the article is out of date, the Met answers with what it says; keeping the knowledge base current is your team's job.
-
It does not link the WhatsApp contact with the website one on its own. If the person gives no name or email, the Met has nothing to look them up with in Crisp.
-
It does not ask for confirmation in the chat before leaving the note or changing segments: if the action is on, it runs it. The admin decides which writes stay on.
-
It does not move the WhatsApp conversation into Crisp. The internal note leaves the context, but the conversation stays in the channel where it started.
FAQ
Questions about this automation
No. It is an internal note: Crisp shows it only to the workspace operators and does not send it to the person.
The Met searches by title in the knowledge base tree. The skill should tell it which language to answer in on each line.
The MCP token has the same permissions as your workspace's REST API token. In Met you can switch off the actions you do not want it to use.
Keep going
Other automations built on the same tools
Customer service and scheduling
Open a Linear issue from a customer report on WhatsApp and tell them how it is going
How do I make customer reports on WhatsApp arrive in Linear as issues?
Customer service
- Starts:
- when the customer writes
- Steps:
- 5, with 6 real tools
- Needs:
- Linear
Customer service and scheduling
Open a Freshdesk ticket from WhatsApp and tell where it stands
How do I turn a complaint that arrives on WhatsApp into a Freshdesk ticket without copying it by hand?
Customer service
- Starts:
- when the customer writes
- Steps:
- 5, with 10 real tools
- Needs:
- Freshdesk
Customer service and scheduling
Open a Gorgias ticket from WhatsApp and escalate it to the team that solves it
How do I move a complaint that arrives on WhatsApp into Gorgias without copying it by hand or losing what the customer said?
Customer service
- Starts:
- when the customer writes
- Steps:
- 5, with 7 real tools
- Needs:
- Gorgias
Customer service and scheduling
Open a Jira issue from a report that comes in on WhatsApp and answer how it is going
How do I make customer reports on WhatsApp arrive in Jira as issues?
Customer service
- Starts:
- when the customer writes
- Steps:
- 5, with 6 real tools
- Needs:
- Jira y Confluence
Want this automation running on your own data?
We will show it to you with your accounts connected, not with a canned demo.