MCP + Notion: Build a Self-Updating Client CRM with AI

Why Most Local Businesses Skip the CRM

What You Are Building

1. A client finishes an appointment. Your booking system (or a simple form) sends appointment data to a webhook.
2. A Python script receives that data and passes it to Claude via the API.
3. Claude reads the appointment notes, extracts key details, and uses the Notion MCP server to update (or create) the client record in Notion.
4. Scheduled automations run daily: Claude checks for upcoming follow-up dates, drafts reminder messages, scans for birthdays, and generates a weekly revenue summary.

• Notion account (free tier works fine)
• Claude API key (Anthropic Console — light usage starts at roughly $5/month)
• Python 3.10+
• A way to trigger webhooks from your booking system (most platforms support this natively)

Setting Up Your Notion CRM Database

{
  "Client Name": {
    "title": [{ "text": { "content": "Sarah Chen" } }]
  },
  "Phone": { "phone_number": "+1-555-842-1923" },
  "Email": { "email": "sarah.chen@email.com" },
  "Last Visit": { "date": { "start": "2026-05-28" } },
  "Services": {
    "multi_select": [
      { "name": "Highlights" },
      { "name": "Blowout" }
    ]
  },
  "Visit Notes": {
    "rich_text": [{
      "text": {
        "content": "Requested balayage with warm tones. Loves the result. Sensitive scalp — use gentle developer. Prefers appointments after 3pm."
      }
    }]
  },
  "Next Follow-Up": { "date": { "start": "2026-07-23" } },
  "Lifetime Value": { "number": 1240 },
  "Status": { "select": { "name": "VIP" } },
  "Total Visits": { "number": 14 }
}

Installing the Notion MCP Server

{
  "mcpServers": {
    "notion": {
      "command": "npx",
      "args": ["-y", "@notionhq/notion-mcp-server"],
      "env": {
        "OPENAPI_MCP_HEADERS": "{\"Authorization\": \"Bearer YOUR_NOTION_INTEGRATION_TOKEN\", \"Notion-Version\": \"2022-06-28\"}"
      }
    }
  }
}

The Four Core Automation Workflows

Workflow 1: After-Appointment Log

import anthropic
import json
from flask import Flask, request, jsonify

app = Flask(__name__)
client = anthropic.Anthropic(api_key="YOUR_ANTHROPIC_API_KEY")

NOTION_DATABASE_ID = "YOUR_DATABASE_ID"

SYSTEM_PROMPT = """You are a CRM assistant for a local service business.
When given appointment data, you:
1. Search the Notion CRM database for the client by name or phone number
2. If found, update the existing record with new visit details
3. If not found, create a new client record with all available fields
4. Always update: Last Visit date, Services, Visit Notes, Total Visits (increment by 1)
5. Calculate Next Follow-Up date based on service type:
   - Haircut: 6 weeks from today
   - Highlights or Color: 8 weeks from today
   - Full Groom (dog): 4 weeks from today
   - Bath only: 3 weeks from today
   - Fitness or personal training session: 1 week from today
6. Add today's service cost to the existing Lifetime Value
7. Set Status to VIP if Lifetime Value exceeds $500

Respond with a plain-text summary of what was updated or created."""

@app.route("/appointment-webhook", methods=["POST"])
def handle_appointment():
    data = request.json

    appointment_context = f"""
New appointment completed:
Client: {data.get('client_name')}
Phone: {data.get('phone')}
Email: {data.get('email', 'Not provided')}
Services today: {', '.join(data.get('services', []))}
Staff member: {data.get('staff_member')}
Appointment date: {data.get('appointment_date')}
Duration: {data.get('duration_minutes')} minutes
Cost today: ${data.get('cost')}
Staff notes: {data.get('notes', 'None')}

Please update the Notion CRM with this appointment information.
"""

    response = client.messages.create(
        model="claude-opus-4-5",
        max_tokens=1024,
        system=SYSTEM_PROMPT,
        messages=[{"role": "user", "content": appointment_context}]
    )

    return jsonify({
        "status": "updated",
        "client": data.get('client_name'),
        "details": response.content[0].text
    })

if __name__ == "__main__":
    app.run(port=5001)

Workflow 2: Follow-Up Scheduler

import anthropic
import schedule
import time
from datetime import date

client = anthropic.Anthropic(api_key="YOUR_ANTHROPIC_API_KEY")

def run_followup_check():
    today = date.today().isoformat()

    prompt = f"""
Today is {today}. Search the Notion Client CRM database and find all clients where:
- The "Next Follow-Up" date is on or before today
- Status is "Active" or "VIP"

For each client found, draft a personalized follow-up SMS message that:
- Uses their first name only
- References their last service (pull from Services field)
- Suggests booking without being pushy
- Stays under 160 characters

Format your output as a numbered list:
1. [Client Name] — [Draft SMS]

If no clients need follow-up today, say so clearly.
"""

    response = client.messages.create(
        model="claude-opus-4-5",
        max_tokens=2048,
        messages=[{"role": "user", "content": prompt}]
    )

    print(f"=== Follow-Up Report: {today} ===")
    print(response.content[0].text)
    print("=" * 40)

schedule.every().day.at("08:00").do(run_followup_check)

while True:
    schedule.run_pending()
    time.sleep(60)

Workflow 3: Birthday and Anniversary Reminders

def run_birthday_check():
    today = date.today()
    month_day = today.strftime("%m-%d")

    prompt = f"""
Today's month and day is {month_day} (ignore the year when checking birthdays).

Search the Notion Client CRM database for any clients whose Birthday field 
matches today's month and day.

For each birthday client found, produce two things:
1. A warm SMS under 160 characters offering 15% off their next visit, 
   mentioning their name and the specific service they last had
2. An email version with a subject line and 3-4 sentence body

If no birthdays today, check for clients whose first-ever visit 
(estimate from the date pattern in notes if available) was exactly 
1 year ago this month — draft a "one year anniversary" loyalty message for those.

List each client separately with clear SMS and Email sections.
"""

    response = client.messages.create(
        model="claude-opus-4-5",
        max_tokens=2048,
        messages=[{"role": "user", "content": prompt}]
    )

    print(f"=== Birthday/Anniversary Check: {today} ===")
    print(response.content[0].text)

schedule.every().day.at("09:00").do(run_birthday_check)

Workflow 4: Monthly Revenue Reporting

from datetime import datetime, timedelta

def generate_monthly_report():
    first_of_this_month = datetime.now().replace(day=1)
    last_month_end = first_of_this_month - timedelta(days=1)
    last_month_start = last_month_end.replace(day=1)
    last_month_label = last_month_end.strftime("%B %Y")

    prompt = f"""
Generate a monthly client revenue and retention report for {last_month_label}
using the Notion Client CRM database.

Please calculate and clearly report:

1. REVENUE SUMMARY
   - Total revenue from all visits last month (sum of Lifetime Value increments)
   - Average spend per visit
   - Highest single-visit spend

2. TOP CLIENTS
   - Top 5 clients by total spend last month

3. CLIENT ACTIVITY
   - Total unique clients served
   - New clients (first visit ever) vs returning clients
   - Clients who visited 3+ times (highly loyal segment)

4. SERVICE BREAKDOWN
   - Which services generated the most revenue

5. RETENTION RISK
   - Clients who visited last month but whose Next Follow-Up date has now passed
   - These are the people most at risk of churning — list their names

6. MONTH-OVER-MONTH TREND
   - Compare to the prior month if data is available
   - Call out any significant changes

Write this as a readable business narrative, not just a data dump.
Start with a one-sentence executive summary.
"""

    response = client.messages.create(
        model="claude-opus-4-5",
        max_tokens=3000,
        messages=[{"role": "user", "content": prompt}]
    )

    print(f"=== Monthly Business Report: {last_month_label} ===")
    print(response.content[0].text)

# Run on the 1st of every month at 7am
schedule.every().day.at("07:00").do(
    lambda: generate_monthly_report() if datetime.now().day == 1 else None
)

Real Example: A Pet Grooming Business with 200 Clients

• Average client visits per year: 3.1
• Lapsed clients (no visit in 10+ weeks): estimated 35–40 at any time
• Time spent on follow-up reminders per week: 2–3 hours of manual texting
• Monthly revenue visibility: "roughly I think around X" level accuracy

• Follow-up rate went from ad hoc to 100% automated. Every pet gets a follow-up text at exactly the right interval for their coat and service type (bath-only dogs get 3-week nudges; full-groom clients get 4-week messages). Average visits per client climbed from 3.1 to 4.7 in the first 6 months.
• Lapsed clients recovered. In the first 30 days, Claude identified 23 clients who had not booked in over 8 weeks. The owner sent the AI-drafted re-engagement messages. Nine rebooked within a week — roughly $720 in recovered revenue with 10 minutes of effort.
• Birthday messages converted at 31%. Fourteen clients received birthday messages in the first month. Four booked a birthday bath package the same week. That is a 28% conversion rate on a zero-cost campaign.
• Time saved: approximately 45 minutes per day that had previously been spent on manual follow-up reminders and memory-jogging before client calls.
• Total monthly cost: $11.40 in Claude API calls for 200 active clients with daily automations.

System Comparison: What Are Your Options?

FAQ

Related Articles

• What Is MCP (Model Context Protocol)? A Plain-English Guide for Local Business Owners
• Best MCP Servers for Local Business: The Complete Setup Guide
• Claude + MCP + Google Calendar: Automate Your Booking System
• MCP Slack Connector: AI Business Alerts in Your Team Channel
• MCP + Supabase: Build an AI Agent with a Real Database Backend