By playing, you agree to uListen's Terms and Privacy Policy and YouTube's Terms of Service.

How this PM uses MCPs to automate his meeting prep, CRM updates, and customer feedback synthesis

Reid Robinson, Principal AI Product Strategist at Zapier, shares how he uses Model Context Protocols (MCPs) to automate tedious tasks and create powerful workflows.

Claire VohostReid Robinsonguest

Feb 2, 202640mWatch on YouTube ↗

Summaries, chapters, transcripts, and FAQs on this page are made by uListen, not YouTube.

CHAPTERS

Jump to publisher timestamps

Chapter summaries

MCPs demystified: connectors that give AI access to knowledge + actions
Claire and Reid open by reframing MCPs in practical terms: app integrations (connectors) for AI tools. They emphasize the two core values—letting an AI read knowledge stored across your apps and take actions inside those apps.
- •Don’t get hung up on the term “MCP”; think “AI app connectors”
- •Two main benefits: access to data in your apps and ability to act in your apps
- •Applies across chat clients and IDEs (Claude, ChatGPT, Cursor, etc.)
- •Why MCPs feel hyped yet underutilized: conceptual complexity
Zapier’s MCP approach: 8,000 apps exposed as a configurable toolset
Reid explains how Zapier exposes thousands of app actions/searches through its MCP server approach. The key differentiator is treating Zapier as a platform to create multiple MCP servers—each with a tailored set of tools for different contexts.
- •Zapier MCP exposes ~8,000 apps and tens of thousands of actions/searches
- •Create curated “collections of tools” for specific workflows
- •Multiple servers possible: different tool bundles for Claude vs. ChatGPT vs. Cursor
- •Goal: give AI apps a single endpoint with all needed tools
On-screen walkthrough: building a custom MCP server with scoped permissions
They walk through configuring a Zapier MCP server by selecting apps (Slack, Evernote, Glean, Coda, Google Calendar) and restricting access to specific notebooks/databases. This highlights how scoping and customization can make tool use safer and more reliable.
- •Add tools from many apps into one MCP server
- •Restrict tools to specific resources (e.g., Coda docs, Evernote notebooks)
- •One URL/connector in Claude can represent many underlying tools
- •Practical benefit: reduces the overhead of managing many separate MCPs
Tooling UX pitfalls: competing tools, naming collisions, and priority control
Claire flags a real-world issue for advanced users: tool-calling ambiguity when multiple MCPs overlap in functionality or naming. They discuss the need for better meta-controls (like priority or disambiguation) as MCP adoption grows.
- •Tool competition can cause models to call the wrong connector/tool
- •Naming collisions (“projects,” “search,” etc.) are common
- •MCP clients need better granular controls and tool priority
- •As MCP ecosystems grow, orchestration/abstraction becomes critical
Claude Projects as a control plane: instructing tool order and field mapping
Reid shares a key tactic: use Claude Projects not just for knowledge, but to enforce detailed tool-usage instructions. By specifying sequences, data destinations, and record creation rules, the same toolset becomes far more consistent and useful.
- •Claude Projects can encode “how to use tools” step-by-step
- •Specify tool order, data placement, and record creation behavior
- •Improves multi-step execution and reduces tool-selection errors
- •Useful for CRM workflows with custom fields and non-standard schemas
Post-meeting notes automation: from transcript to Coda/CRM updates
Reid demonstrates a post-meeting workflow that turns meeting artifacts (e.g., Granola notes) into structured updates. The system checks whether records exist, searches internal sources (like Glean), and then creates or updates entries with next steps and opportunity details.
- •Automates tedious post-meeting logging and follow-ups
- •Checks for existing records, then enriches via internal search/lookup
- •Writes structured outputs to Coda and optionally logs activities in HubSpot
- •Custom instructions teach the model how to fill unique CRM fields
Deterministic workflows vs. agentic/MCP instructions: reliability tradeoffs
Claire and Reid compare Zapier’s traditional step-by-step workflow builder with more agentic, instruction-led tool use. Reid notes deterministic flows can handle long-running tasks better today, while MCP shines at “meeting users where they work” inside their preferred AI apps.
- •Deterministic workflows can support longer, asynchronous processes
- •MCP interactions are time-bounded; long multi-minute lookups may fail
- •Agentic instructions reduce manual branching logic (“if record exists…”)
- •Biggest unlock: tool access embedded in the AI surfaces people already use
Idea jammer + enterprise rollout: role-based toolsets for every employee
Reid briefly introduces an ‘idea jammer’ project that uses tools and prompting methods to challenge and develop ideas. They also discuss enterprise adoption: admins want preconfigured tool bundles so employees can connect once and immediately have the right tools for their role.
- •Dedicated projects can support creative/strategic workflows (“idea jammer”)
- •Projects can include prompting methods to challenge thinking
- •Enterprises want standardized, role-appropriate tool bundles
- •Ops/admin-managed defaults reduce setup burden and increase adoption
Customer interview prep workflow: automated context before meetings
Reid shows a more deterministic Zapier workflow for interview preparation, triggered on upcoming meetings. It fetches internal customer/company data, processes it into a readable brief, and appends it to a Coda page so he never joins a call without context.
- •Solves the “I don’t know who you are” problem in interviews
- •Fetches company/user info from internal systems (e.g., Databricks outputs)
- •Summarizes and appends a prep brief into Coda for the meeting
- •Improves confidence and professionalism for PMs, Sales, and PMMs
Why Gemini is used here: better at file-based inputs (HTML/PDF)
They explain the model choice: Gemini performs especially well when the pipeline involves files like PDFs or converted HTML documents. Reid converts HTML to a file for efficiency and lower token usage, reinforcing the importance of matching model strengths to data formats.
- •Gemini is preferred for file-heavy workflows
- •Internal data output comes as HTML/PDF-like artifacts
- •Converting HTML to a file can reduce token usage and improve processing
- •Practical tip: choose models based on input modality, not brand loyalty
Customer feedback synthesis: searchable internal bot + trend surfacing
Reid describes systems for synthesizing and operationalizing feedback at scale. They analyze feedback, create review pages in Coda, and provide internal teams (Sales, PMM, Design) a quick way to search what users are reporting and requesting.
- •Automated analysis identifies trends and produces Coda pages for review
- •Makes feedback searchable for non-core teams (Sales/PMM/Design)
- •Supports targeted queries (e.g., errors, specific workflows, recent issues)
- •Extends beyond the build team to organization-wide enablement
Closing the loop: auto-proposed FAQs from tickets/transcripts with human approval
A standout workflow: after a support ticket closes or a chatbot transcript completes, AI extracts the core question, solution, and checks whether the knowledge base already covers it. If not, it proposes a new FAQ entry that a human reviews and approves, keeping the bot’s knowledge fresh.
- •Analyze closed tickets/transcripts to extract FAQ candidates
- •Detect gaps vs. existing knowledge base content
- •Human-in-the-loop approval step before publishing
- •Creates a “virtuous cycle” that improves support speed and quality
Brainstorming framework: “run ChatGPT in your sleep” + Mad Libs discovery
They discuss ideation techniques for finding high-leverage AI automations. Reid offers the prompt, “If you could run ChatGPT in your sleep, what would you do?” and shares an experiment using Mad Libs-style guided prompts to uncover pain points and use cases.
- •Use “AI in your sleep” framing to find always-on automations
- •Mad Libs-style UX can help users articulate needs and workflows
- •Good discovery prompts expose pain points and desired outcomes
- •Focus expands from speed to quality improvements
Recap of the operating system: MCP toolsets, Projects, meeting prep, and feedback loops
Claire summarizes the main takeaways: MCPs as custom tool bundles, Claude Projects for precise tool instructions, and workflows that reduce customer-facing drudgery. The emphasis is on fewer tabs, better preparation, and continuously improving customer knowledge.
- •Zapier MCP enables curated tools accessible inside AI clients
- •Claude Projects improves tool-calling reliability and sequencing
- •Key wins: CRM updates, meeting context, feedback-to-FAQ loops
- •Overall outcome: higher leverage customer-facing work
Personal use cases: family calendar via photo + kids’ songs with Suno; NotebookLM interview prep
In a lightning round, Reid shares personal automations: turning a physical family calendar photo into Google Calendar events via Claude + MCP, and using Suno to create custom kids’ songs (and teach prompting). He also explains how NotebookLM audio overviews helped his wife prep for interviews with highly personalized, company-specific briefings.
- •Family calendar: photo → Claude Project → Google Calendar updates (with travel buffers)
- •Suno: generate personalized kids’ songs; playful way to learn prompting
- •Music as memorability: even turning training transcripts into songs
- •NotebookLM: tailored audio interview prep from job/career pages and research

Publisher timestamps

Get more out of YouTube videos.

Read along on your phone, search it beside the video, or ask your favorite AI app what was said.