Configure MCP Registries
Registries are how teams discover and govern MCP servers at scale. Learn the difference between the public MCP catalog, the GitHub MCP Registry and a private/enterprise registry — and how each one feeds into a host's allow-list pipeline.
Configure MCP Registries
A registry is a catalog of MCP servers. It tells you which servers exist, what they expose, and how to install them. It does not run them, and it does not gate them — those remain the host's responsibility. Picking the right registry is the difference between a curated, governable agent stack and a wild-west install spree.
The three registries you will meet on the exam
| Registry | Audience | Strength | Use it when | | --- | --- | --- | --- | | Public MCP catalog (modelcontextprotocol.io ecosystem) | Open community | Largest selection | You are evaluating and the workload is non-sensitive. | | GitHub MCP Registry | GitHub Copilot users | Curated, partner-vetted | You want Copilot-friendly servers with a baseline of review. | | Private / enterprise registry | Internal devs | Allow-list at the discovery layer; SSO-aware | You are at an organisation that has a security review process. |
Registry → host → consent
Even with a perfect registry, the host's permission model still runs. A pipeline that takes safety seriously looks like this:
- Registry filters what is discoverable.
- Host config filters what is installed.
- Toolset flags filter what is exposed.
- Consent gates filter what is executed.
The exam likes to ask "which control would have prevented this?" Always pick the layer closest to the failure. A registry cannot stop a runaway tool call; only a consent gate can.
Build a Copilot MCP entry
Assemble the config
Assemble a VS Code-style MCP entry that pins a server discovered from a private registry.
- Empty — pick from the right.
Where this shows up on the exam
If the question gives you a multi-tenant enterprise scenario, the answer almost always includes "private registry" + "pinned versions" + "host consent enforced". If it gives you a single-developer scenario on a personal machine, public catalog + per-call consent is enough.
Key terms
- MCP registry
- A discoverable catalog of MCP servers with metadata about their tools, install instructions and provenance.
- GitHub MCP Registry
- A GitHub-hosted curated list of MCP servers from partners and the community, used to discover servers vetted for Copilot.
- Private registry
- An enterprise-controlled MCP catalog (often self-hosted) that only lists servers approved for internal use, behind SSO.
- Pinning
- Locking a server to a specific version or content hash so an unexpected upstream change cannot silently alter the agent's tool surface.
Common pitfalls
- Treating any MCP server you find on the public registry as 'safe by default' — registries are catalogs, not security audits. Provenance still matters.
- Letting servers auto-update from the registry on every host launch; one bad release will silently change the agent's tools and prompts under your feet.
- Skipping pinning so an attacker who compromises a server's npm package immediately attacks every agent that installed it via `npx -y`.