Most MCP servers don't need to exist. Your case might be an exception.—Martian Chronicles, Evil Martians’ team blog
This title could be clearer and more informative.Try out Clickbait Shieldfor free (5 uses left this month).
Most MCP servers being built today are premature product interfaces that don't need to exist. Before building one, teams should ensure their API is shaped around user intent (not database rows) and their CLI is agent-operable. A skill — instructions and playbooks that teach an agent to use existing capabilities — is often a cheaper first step. The primary signal that justifies an MCP server is cross-client demand: AI clients you don't control need access to the same operations. The post provides a decision framework with eight questions, details architectural amplifiers that sharpen the case, and catalogs hidden costs including context budget consumption (55K+ tokens before first message in some setups), operational overhead, security debt (88% of audited MCP repos have credential issues), approval laundering risks, and client fragmentation across tools like Cursor, GitHub Copilot, and Gemini CLI. Common failure modes include API dumps, tool overload, missing security models, no evals, and no versioning story.
