From Experiment to Enterprise: How Block Operationalized MCP at Scale
summary
Block deploys MCP organization-wide in about two months, building an open-source MCP client called Goose that connects to Snowflake, Jira, PagerDuty and other MCP servers; internal rollout includes automatic installs, SSO short‑lived credentials, and an allow list. Engineering reports 75% of Goose users save 8–10 hours weekly, internal catalog grows to 100+ MCP servers, and non‑engineers use natural‑language queries and shared “recipes” to deploy prototypes and automate workflows. Model Context Protocol and DataBricks model hosting enable model choice and failover.
AI-written summary · about
chapters (12)
- 0:00Introduction and context
- 3:01Developing Goose and MCP
- 6:28Scaling Goose for all employees
- 9:17Overcoming technical hurdles
- 13:18Securing and managing MCP servers
- 15:25Support and adoption strategies
- 17:11Education and enablement efforts
- 18:37Real-world use cases at Block
- 21:09Evolving Goose’s features
- 23:56AI Champions program
- 26:48AI-native engineering culture
- 29:17Closing and future outlook
related talks
29mMCP in 2026: Context is All You Need
44mSupercharging DevOps with MCP (Without Opening a Security Hole)
18mMCP: Origins and Requests For Startups
33mThe Model Context Protocol: Connecting AI to Everything
47mMCP Demystified: Cutting Through the Hype-What It Is, What Isn't & Its Role in the Agentic RAG Stack
30mComposable AI: Building Next-Gen AI Agents with MCP
46mMCP for LLM Agents, Infrastructure APIs & Data Services
35mMCP in Action: Real-World Use Cases & Design Patterns