What is MCP and Why Should You Care?
Model Context Protocol (MCP) is an open standard that lets AI tools connect to external systems - databases, APIs, file systems, SaaS tools - through a unified interface. Think of it as USB-C for AI: one standard connector that works with everything.
If your team is using Claude Code, Cursor, or any MCP-compatible AI tool, setting up MCP servers is the single highest-leverage thing you can do to make AI useful in your daily workflow. We cover how Claude Code transformed our own engineering process in How Claude Code Changed Our Workflow.
The Before and After
Before MCP: You copy-paste context from Slack, GitHub, and your docs into a chat window. The AI gives you a generic answer based on incomplete information. You do this fifty times a day.
After MCP: Your AI tool has direct, read-only access to your repos, tickets, docs, and internal APIs. It answers questions grounded in your actual data. You ask "what is the status of the auth refactor?" and it checks GitHub PRs, Linear tickets, and your Notion docs to give you a real answer.
MCP Servers We Deploy Most Often
GitHub MCP Server
This is the one everyone should start with. It gives your AI tool access to repos, PRs, issues, and code search.
Use cases:
- "Summarize the changes in the last 5 PRs on main"
- "Find all TODO comments in the payments module"
- "What issues are assigned to me this sprint?"
Setup complexity: Low. The official GitHub MCP server works out of the box with a personal access token.
Notion MCP Server
If your team lives in Notion (and many of our clients do), this is transformative. Your AI tool can search across all your docs, databases, and wikis.
Use cases:
- "What did we decide about the pricing model in last week's meeting notes?"
- "Find the onboarding checklist for new engineers"
- "What are the open action items from the Q1 planning doc?"
Database MCP Server (Postgres/MySQL)
For engineering teams, connecting your AI tool to a read-only replica of your database is incredibly powerful.
Use cases:
- "How many users signed up in the last 30 days?"
- "What is the schema for the orders table?"
- "Write a query to find customers with more than 3 failed payments"
Important: Always connect to a read-only replica. Never give AI tools write access to production data.
Custom Internal API Servers
This is where MCP gets really interesting. We have built custom MCP servers that connect to:
- Internal deployment dashboards
- Customer support ticket systems
- Analytics platforms
- CI/CD pipelines
The pattern is always the same: wrap your internal API in an MCP server, define the available tools and their schemas, and your AI tool can now interact with it.
Security Considerations
MCP servers are powerful, which means security matters:
- Principle of least privilege. Every MCP server should have the minimum permissions needed. Read-only where possible.
- Token management. Use short-lived tokens or OAuth where supported. Never hardcode secrets.
- Network isolation. MCP servers that access internal systems should run inside your VPC, not on developer laptops.
- Audit logging. Log every MCP request so you know what your AI tools are accessing.
Getting Started
Our recommended starting path:
- Week 1: Set up the GitHub MCP server for your engineering team. This is low-risk, high-reward.
- Week 2: Add Notion or your primary docs platform.
- Week 3: Evaluate which internal systems would benefit from MCP access and scope a custom server.
Each step builds on the last, and your team gets incrementally more value from their AI tools.
If you want help setting up MCP servers for your team, or if you need custom MCP servers for internal systems, reach out or see our full AI agent deployment service. We have built dozens of these and can get you running in days, not weeks.