Case Study: Block

Overview

Field	Value
Company	Block, Inc.
Industry	Fintech / Payments
Headquarters	San Francisco, CA
Deployment Date	Early 2025
MCP Version	Compatible with Anthropic MCP spec
Enterprise Readiness Score	⭐⭐⭐⭐⭐

Use Case

Block deployed its internal AI agent Goose company-wide on top of MCP. The goal was to give all employees — not just engineers — access to an AI agent capable of querying internal systems, automating workflows, and reducing time on repetitive tasks.

Usage expanded beyond the engineering org to include design, product, support, risk, data, and operations teams.

Connected Systems

System	Type	Direction	Auth Method	Notes
GitHub	Developer Tools	Read/Write	OAuth	Scoped to repo-level permissions
Jira	Project Management	Read/Write	API Token	Included in approved dev bundle
Slack	Messaging	Read	Bot Token	Message-read only; no admin permissions
Snowflake	Data Warehouse	Read	Service Account	Read-only by policy; write requires approval
Google Drive	Document Storage	Read/Write	OAuth	Approved productivity bundle
Internal APIs	Proprietary	Read/Write	Internal Auth	Custom MCP servers written in-house

Architecture Pattern

Sandboxed Developer + Hub-and-Spoke

Block writes its own internal MCP servers and bundles approved servers for distribution. Access is managed through risk-tiered tool behavior restrictions.

Governance Controls

Access management: Approved MCP server bundles distributed per role/team
Risk classification: Tool behaviors restricted by risk level
Approval gates: Write operations to sensitive systems require elevated approval
Data policies: LLM data access routed through enterprise-managed infrastructure with data-use protections

Outcomes

Thousands of employees use Goose daily across all departments
Many users report saving 50–75% of time on common tasks
Some multi-day projects compressed to hours
Non-technical teams (support, design, risk) independently run data queries and workflows

Source Links

Evidence Quality Notes

All systems, governance controls, and outcome metrics are directly stated in the published Goose blog post. This is the strongest publicly documented enterprise MCP deployment as of March 2026.

Map Your MCP Stack

Your MCP Servers

Closest Reference Architectures