How Block scaled MCP across 12,000 employees in two months

Removing friction to reach 15 job functions beyond engineering.

Most companies are still experimenting with AI agents while Block already has 12,000 employees using them across 15 different job functions. In her presentation at All Things Open, Angie Jones, VP of Engineering at Block, shares how the company operationalized Model Context Protocol (MCP) at scale, turning an internal developer tool called Goose into a general-purpose agent that non-engineers could actually use without friction.

Block started with Goose, a development agent that could complete tasks, not just generate code. When they aligned with the Model Context Protocol, they rewrote Goose as an extensible MCP client. Within a month, 75 percent of engineers were saving 8 to 10 hours a week. But the real challenge came when non-technical teams wanted access. People couldn’t install MCP servers, didn’t understand API keys, and had no idea where to find the tools they needed.

Block solved this by removing friction at every level. They added Goose to their internal software center for auto-installation and auto-updates. They built over 100 internal MCP servers bundled by default instead of making people hunt for external ones. They implemented OAuth with identity provider integration so employees saw a familiar SSO flow instead of managing credentials. They added dynamic context management to automatically enable and disable servers based on user queries, plus a context summarizer for longer conversations.

The payoff was immediate. A fraud analyst detected suspicious transaction patterns. An employee analyzed 80,000 sales leads in one hour instead of days. Someone with zero technical background used Goose to build a functional internal tool and asked where to put it on GitHub, only to find she didn’t even have a GitHub account because she wasn’t a developer. That moment captured what Block achieved, turning AI accessibility into a reality across the entire company.

Read more: Discover Goose: Automate your developer setup with this AI agent

Key takeaways

• Remove friction at every step. Auto-installation, bundled servers, OAuth flows, and dynamic context management each directly drove adoption across different roles.
• Build internal infrastructure instead of relying on external tools. Block created their own MCP servers for security and control, not distrust of open source options.
• Community support drives adoption faster than mandates. Slack channels and consistent education unlocked usage across 15 job functions in just two months.

Block didn’t just adopt MCP, they operationalized it end-to-end. The result is a general-purpose agent helping fraud analysts, salespeople, customer service teams, and engineers work faster, all within two months. Watch the full talk to learn how they removed friction at scale, the pain points they encountered, and the early decisions that accidentally unlocked adoption across the entire company.

More from We Love Open Source

• Deep dive into the Model Context Protocol
• Discover Goose: Automate your developer setup with this AI agent
• Why AI agents are the future of web navigation
• What if your AI agent could actually help?
• AI agents are here—are your skills ready?