← All case studies

AI agents · 2026

MetaMCP as a Cloud Run MCP gateway

How a local MCP gateway becomes a cloud-facing tool surface for agents on Cloud Run.

3 min read

MetaMCP as a Cloud Run MCP gateway

Local MCP is useful.

Remote MCP is a system boundary.

MetaMCP started as a way to reduce many MCP tools into a smaller tool surface. The next question was practical: can that surface live behind Cloud Run, get registered for agents, and leave evidence after tool calls?

The problem

Most MCP work begins on a laptop.

That is where the idea gets proven.

The production question is different.

Can another agent call it from a cloud runtime?

Can the endpoint be registered?

Can auth, secrets, tests, and evidence sit around the tool call?

That is the part worth building.

The system shape

MetaMCP keeps the local developer path intact.

Then it adds the parts needed for a remote MCP surface:

What this proves

Agent question System answer
Where are the tools? Behind a smaller MetaMCP gateway.
How does a cloud agent call them? Through Streamable HTTP on Cloud Run.
How are tools registered? With Agent Registry metadata and a manual toolspec.
How does auth work? Cloud Run caller identity plus app-level bearer auth.
How do we know what happened? Logs, smoke tests, prompt fixtures, and Mentu evidence.

Why it matters

Agents get interesting when they stop living in a demo window.

Then the boring parts become the product:

MetaMCP is a small open-source way to show that path.

Related writing

Artifacts