Question 1

How would you explain MCP to a non-technical manager?

Accepted Answer

Start with the analogy, then move to business value. "Our AI assistant can write code but doesn't know which bug it is writing for, because it has no access to Sentry or Jira. MCP is the standard 'socket' that grants that access. Because the socket is standard, you don't build a separate cable for every system." Business value: Developers stop copying data by hand — time saved, transcription errors gone. We connect our internal systems once and they work across tools. The standard is open, so we aren't locked to one vendor. Name the risk too (it earns trust): an MCP server grants the agent access to real systems, so permissions and audit matter. Present that as a security decision, not a technical one.

Question 2

Which MCP server would you connect first for your team, and why?

Accepted Answer

Justify the pick with a criterion: the data that gets copied by hand most often. On an Android team that is usually crash reporting (Sentry, Crashlytics). Because: Stack traces are long and easy to copy wrong. It is high-value context for the agent — it points straight at the code. It is read-oriented, so risk is low: the server supplies data, it doesn't change the system. Second would be issue tracking (Jira/Linear), since ferrying requirements to the agent is daily work. Then add an important principle: I don't connect a write-capable server first. I start with read-only servers and let the team settle into the workflow; writes (changing ticket status, triggering a deploy) are a later and separate decision.

Question 3

What is the alternative to connecting MCP, and when is the alternative better?

Accepted Answer

List the alternatives and draw the boundary — not every problem needs MCP. Alternatives: 1. A CLI command — if the system has its own CLI (gh, firebase, adb) the agent can just run it. Simple, no new component. 2. A skill plus a script — call the script from inside a skill. 3. Copying by hand — perfectly fine for rare cases. When MCP is better: Queries are frequent and take many shapes (writing a CLI command per variant is painful). The result is structured and the agent needs to process it. Several tools want access to the same system. Authentication is complex (OAuth) and should be centralised. When the CLI is better: when the system is already driven by a CLI and the commands are simple. Writing a dedicated MCP server for adb is usually unnecessary — the agent can run adb directly. Close with the principle: MCP adds a component, and a component is maintenance cost. Add it only when it pays for itself.

Without MCP	With MCP
Every tool writes a separate integration for every system	Every system exposes one MCP server, once
The integration is tool-specific — rewritten when the tool changes	The server works with every MCP client
Waiting on the tool vendor to connect your internal system	You write your own server yourself

What MCP is and what it solves

📚 Sources and documentation