How this PM uses MCPs to automate his meeting prep, CRM updates, and customer feedback synthesis | Reid Robinson (Zapier)

What It Covers

Reid Robinson, Product Manager at Zapier, demonstrates how Model Context Protocol servers transform AI tools into automated assistants for customer-facing work. He shows practical implementations using Claude Projects with Zapier's 8,000-app MCP platform to automate CRM updates, meeting preparation, and customer feedback synthesis while maintaining data quality through human-in-the-loop workflows.

Key Questions Answered

• •MCP Implementation Strategy: Think of MCPs as app integrations for AI tools rather than technical infrastructure. Create custom collections of tools from apps like Slack, Google Calendar, and CRM systems, then connect them to Claude, ChatGPT, or Cursor through a single URL. Zapier's MCP platform exposes 30,000 actions across 8,000 apps, letting users build role-specific tool sets that work automatically when employees log in.
• •Claude Projects for Tool Orchestration: Use Claude Projects to provide detailed instructions on which MCP tools to use, in what sequence, and how to populate specific fields. This approach solves the tool-calling priority problem where models struggle to select the right tool. Create separate projects for different workflows like CRM updates or daily planning, each with customized instructions that dramatically improve execution accuracy across multiple sequential tools.
• •Automated CRM Workflow: Build a post-meeting workflow that searches internal databases for existing customer records, looks up company usage data through internal tools like Glean, and creates or updates CRM entries with meeting notes and next steps. Train the model on your specific CRM fields and custom choices since every organization uses unique configurations. This eliminates the manual task customer-facing teams universally hate.
• •Model Selection by Data Type: Choose Google Gemini models specifically for file-based workflows, particularly PDFs and HTML documents. Gemini handles large file contexts more effectively and uses fewer tokens than OpenAI or Anthropic models. For customer research workflows that pull HTML data from internal systems, convert to files before processing to optimize both performance and cost while maintaining output quality.
• •Self-Improving Knowledge Base: Create a system where closed support tickets and chatbot transcripts automatically analyze conversations to identify core FAQs, compare against existing knowledge bases, and propose new entries for human review. Approved entries feed directly into customer-facing chatbots. This maintains up-to-date help content without manual quarterly reviews, improving answer quality while operating at scale that manual teams cannot achieve.