Lesson 5: Build by talking to Claude

SRSkyline RealtyLesson 5 of 6 · 12 min

Goal: Arjun describes a workflow in plain words; Claude builds it through MCP.

Arjun asks ClaudeClaude uses Auto Appy's MCP toolsA Draft workflowArjun reviews and goes Live

You'll learnMCPClaude and ChatGPT

123456

AM

The situation

I think of new automations while I’m driving between sites. I don’t have time to sit at the editor. I want to describe what I need and have it built, and then check it in the evening.

Arjun Mehta, Founder · Skyline Realty

How it works

Auto Appy has an MCP server: a standard socket that AI assistants like Claude can plug into. Claude reads the list of Auto Appy tools (create a workflow, add a step, get the webhook URL, test a step…) and uses them for Arjun, one approved call at a time. New to MCP? What is MCP? explains it in five minutes.

Animated explainerHow an AI assistant uses Auto Appy through MCP

1MCP (Model Context Protocol) is a standard plug that lets AI assistants use apps. Think of USB-C, but for AI.

2Arjun at Skyline Realty asks Claude in plain words: "When a MagicBricks lead comes in, add it to the CRM and WhatsApp the brochure."

3The assistant asks Auto Appy's MCP server what it can do. Auto Appy answers with its list of tools.

4The assistant uses the tools it needs, one at a time: create the workflow, then add the steps. You approve each one.

5Auto Appy does the work and replies. The assistant tells you the workflow is ready to test.

Connect Claude (two minutes)

1. In Claude, open Settings → Connectors and choose Add custom connector. Name it Auto Appy, URL https://api.autoappy.com/mcp.
2. Click Connect. Auto Appy opens; sign in as Arjun.
3. On Connect Claude to Auto Appy, tick Read workflows and Build and edit workflows. Leave Run and do things for you off for now. Click Allow access.
4. Start a new chat in Claude.

Prefer a token? Settings → MCP connections → Create connection gives one for clients that don’t support sign-in:

autoappy.com/settings/mcpSample data

A token connection with read and build permissions only.

×

Arjun’s conversation

Arjun: When a MagicBricks lead for Skyline Crest comes in, add it to the CRM in the Premium pipeline, assign it to Karan, and WhatsApp the buyer our crest_welcome template.

What Claude does, step by step (each tool call shows in Claude for approval):

1. autoappy_app_catalog to find the CRM and WhatsApp actions and their fields.
2. autoappy_create_workflow to create a Draft called MagicBricks → Premium.
3. autoappy_get_trigger_setup to get the webhook URL, which Claude gives Arjun to pass to MagicBricks.
4. autoappy_send_test_data with one realistic sample lead, so later steps can map its fields.
5. autoappy_add_node twice (Create lead, Send approved template), then autoappy_list_mapping_fields and autoappy_configure_node to fill the fields.
6. autoappy_list_accounts and autoappy_set_account to attach Skyline’s saved CRM and WhatsApp connections by name.

Claude ends with a link to the new workflow. In the evening, Arjun opens it, tests each step and sets it Live himself.

Stay in control

• Keep Claude’s tool approval on (Needs approval) for anything that changes workflows.
• Give Run and do things for you only when you need it: those calls send real messages.
• Claude never sees passwords or the credentials inside connections; it picks saved connections by name.
• Every call is logged in Settings → MCP connections → Activity, and Revoke disconnects Claude instantly.

autoappy.com/settings/mcpSample data

×

Good prompts

Build

“Build a workflow: when a Facebook lead form lead comes in, create a CRM lead in the Rental pipeline and WhatsApp them rental_welcome. Leave it as a Draft.”

Explain

“Explain what the ‘Reply to every property enquiry’ workflow does, step by step, and add a sticky note with your summary.”

Check

“Which of my workflows had failed runs this week? Show the error for each.”

Change

“In ‘Book and remind site visits’, change the reminder from 2 hours to 3 hours before the visit.”

✍️Try it yourself

1. Connect Claude with read-only permission and ask it to describe each of Skyline’s workflows.
2. Ask it to build Lesson 2’s budget split as a new Draft. Compare its result with yours.
3. Look at Activity in MCP connections. Which tools did it call?

What you learned

• MCPA standard plug that lets AI assistants use Auto Appy's tools.
• Connect ClaudeCustom connector, sign in, choose permissions, Allow access.
• MCP toolsEvery tool, and the permission each one needs.

Next: Plug in the website and a daily report →

🎓 For trainers: how to teach this lesson

• Do it live with the class: let a student dictate the request; you approve each tool call on the projector.
• Talk about trust: what would you let an assistant do on its own, and what always needs a human?
• Compare: building in the editor (Lesson 1) versus by conversation. When is each better?