MCP Gateway
The Model Context Protocol is the open standard for connecting agents to tools, data, and systems through one uniform interface instead of bespoke integrations. As platform infrastructure, the work is making that connectivity enterprise-ready -- auth, access control, and observability -- usually via an MCP gateway.
The Pattern
"Everyone is building MCP servers... They're good demos, but not ready to turn into production." -- Tobin South, WorkOS (source)
The Model Context Protocol (MCP) is the open standard for connecting an agent to the world outside its context window -- tools, data, and systems -- through a uniform interface instead of one-off integrations. An official registry already lists thousands of servers. For the platform, MCP is the connective tissue: it is how every agent in the org reaches internal systems without each team reinventing the plumbing.
The platform problem is making that connectivity enterprise-ready. Anthropic's forward-deployed teams describe a "three-headed hydra" -- observability (who is using which tools), access control (who is allowed to), and security (is a server safe, can untrusted clients reach private data) -- which is why most enterprises struggle to run more than single digits of MCP servers (Karan Sampath, Anthropic). The emerging answer is an MCP gateway: a single root of trust that handles auth, routing, and audit for every server behind it.
Why It Matters
MCP turns an N×M integration problem -- every agent wired to every system -- into N+M against one protocol. But a hacky server is not production: it needs authentication, scoped permissions, audit logs, and agent identity before an enterprise can trust it (Tobin South, WorkOS). Treating MCP as platform infrastructure -- a gateway in front of many servers, with scoped permissions and a registry of what exists -- is what lets tool access scale past the demo. It shares a control point with the model router: one routes model calls, the other routes tool calls.
Sources
Last reviewed: 2026-06-24