Summary
This guide helps you set up Wati MCP and start working with your AI assistant (such as ChatGPT or Claude) to manage your sales and support workflows in Astra. You’ll learn how to initiate your first conversation, test key capabilities, and follow best practices during your first week to get reliable results.
Instructions
Part 1: Set up your first conversation
To begin, you need to provide your AI assistant(Claude or ChatGPT) with clear instructions so it understands how to operate within your Wati workspace.
Step 1: paste your setup instructions
Copy and paste the following into your first message to the AI assistant. You only need to do this once per conversation.
You are helping me operate Wati, the agentic platform for sales and
support on WhatsApp, web, and voice. You have access to my Wati
workspace via MCP.
When I ask you to do something, use the Wati tools to actually do it.
Do not just describe what to do.
When building or modifying an agent, follow Astra's six-phase
lifecycle: Discover, Build, Eval, Deploy, Optimize, Voice. Never
publish an agent below 95% eval score across Intent Match, Quality,
Hallucination, and Escalation.
Ask me clarifying questions before building, not after.
Ready when you are.
Once submitted, the AI assistant will confirm that it is connected and ready.
Step 2: Try these 3 tasks
After setup, test these common workflows to understand what Wati MCP can do.
1. Build your first Astra agent
Type:
Build me a customer support agent for my business. Pull content from [your website URL]. Connect it to my CRM. Deploy it to web first, and then WhatsApp once I’ve tested it.
2. Check performance
Type:
Show me how my agents performed last week. Highlight any conversations where the agent handed off to a human and explain why.
3. Fix a mistake
Type:
Look at conversation [ID]. The agent was too pushy when the customer mentioned price. Update the agent's instructions to lead with empathy on pricing objections, then run an eval to confirm the fix.
Part 2: Best practices for your first week
Follow these tips to improve performance and avoid common issues:
1. Focus on one agent at a time
Start with a single use case, such as:
Refund handling
Lead qualification
Build, test, and refine one agent before creating additional ones. This helps maintain quality and reduces complexity.
2. Use test mode before going live
Each agent includes a Test Mode for safe testing.
Example:
Ask your AI assistant: Send a test message to the agent saying I want to know about your refund policy.
Review how the agent responds before deploying it to customers.
3. Run a weekly check-in
Schedule a quick review at the start of each week.
Ask your AI assistant: How did the agent perform last week? What should I improve?
This helps you identify issues early and continuously improve performance.
4. Always run an eval before deploying
After updating an agent, validate the changes to avoid unintended issues.
Example:
Run an eval to confirm this does not break anything.
Wati will check metrics such as:
Intent match
Response quality
Hallucination risk
Escalation handling
5. Review permissions carefully
Your AI assistant operates with the same permissions as your Wati account.
If you are an admin on a Business or Enterprise plan, you can:
Restrict access to specific features
Control what the AI can view or modify
Use Admin Controls to manage these settings.
6. Connect your existing tools
Wati becomes more powerful when integrated with your existing systems like CRM, e-commerce, or team collaboration tools.
Example:
For every hot lead my Wati agent qualified yesterday, create a follow-up task in HubSpot and post a summary to the #sales channel in Slack.
This allows you to automate workflows across multiple platforms with a single command.
You’re now ready to start using Wati MCP. Begin with a simple use case, test thoroughly, and improve gradually to get the best results.