MCP Registry vs MCP Gateway: Discovery vs Runtime Control

- Share:





3366 Members
An MCP registry and an MCP gateway get mentioned in the same breath, but they answer different questions at different times. A registry answers "which MCP servers exist, who published them, and how do I install or reach them?" It is consulted before a connection is made. A gateway answers "may this user's agent call this tool, with these arguments, right now?" It sits in the path of every tool call, so the agent gets the access the user agreed to and does not simply inherit everything the user can do.
Teams that adopt one and assume it covers the other end up with a gap. This post explains what the official MCP Registry does and doesn't do, based on its own documentation, what a gateway adds at runtime, and how the two fit together.
The official MCP Registry launched in preview in September 2025. Its documentation describes it as the official centralized metadata repository for publicly accessible MCP servers, and still labels it a preview in which breaking changes or data resets may occur before general availability.
A few properties matter for security planning:
server.json document with the server's name, where to get it (an npm, PyPI or OCI package, or a remote URL), how to run it, and a description. The code itself lives in package registries.io.github.alice/weather-server can only be published by someone who authenticated as that GitHub account, and com.example/server requires proving control of example.com through a DNS or HTTP challenge, according to the authentication guide.Here is a real lookup from the registry's API, trimmed. At the time of writing it returns the GitHub MCP server's latest entry:
curl "https://registry.modelcontextprotocol.io/v0.1/servers/io.github.github%2Fgithub-mcp-server/versions/latest"
{
"server": {
"name": "io.github.github/github-mcp-server",
"version": "2.2.0",
"packages": [
{ "registryType": "oci", "identifier": "ghcr.io/github/github-mcp-server:2.2.0", "transport": { "type": "stdio" } }
],
"remotes": [
{ "type": "streamable-http", "url": "https://api.githubcopilot.com/mcp/" }
]
},
"_meta": {
"io.modelcontextprotocol.registry/official": { "status": "active", "isLatest": true }
}
}
That entry tells you who published the server, which version is current, and where to fetch or reach it. It tells you nothing about what a given agent should be allowed to do with it.
The registry is explicit about its limits. Its moderation policy says consumers should assume minimal-to-no moderation. It removes illegal content, malware, spam and non-functioning servers, and it does not remove low-quality servers or servers with security vulnerabilities. Security scanning is delegated to package registries and downstream aggregators. When a server is removed, its status is set to deleted while the metadata stays readable through the API, so anything that syncs from the registry should check that field.
Beyond moderation, a registry has no view of runtime at all:
tools/call request or its arguments.Namespace verification proves where a server came from. It doesn't prove the server is safe, and it doesn't make any particular call appropriate.
An MCP gateway is a proxy between MCP clients and the MCP servers they connect to. Because every request passes through it, it can make and record a decision for each one. A gateway typically handles:
The registry works at the level of server listings, and the gateway works at the level of individual calls. For the difference between a gateway and a plain proxy, see MCP gateway vs MCP proxy.
| MCP registry | MCP gateway | |
|---|---|---|
| Question it answers | Which servers exist and who published them | Whether this call may run now |
| When it acts | Before connection (discovery, install) | On every tool call |
| Unit of decision | A server version | A user, agent, tool and arguments |
| Identity it checks | The publisher's namespace | The end user and the agent |
| Security it provides | Publisher authenticity, basic takedowns | Per-call authorization, consent, audit |
| What it misses | Runtime behavior, users, arguments | Whether a server should be in your catalog at all |
The two controls stack. A workable setup looks like this:
status field on every sync.Some products combine both roles in one MCP gateway registry, so the registration step also decides where the gateway sends traffic. AWS's Security Bulletin 2026-180-AWS shows the risk: in mcp-gateway-registry versions 1.0.3 through 1.34.0, a user allowed to register or edit a server could point its upstream address at internal services (CVE-2026-125926, fixed in 1.35.0). Treat registering and editing servers as a privileged permission, limit the gateway's outbound traffic, and review stored upstream addresses and logs after upgrading. We walk through it in MCP gateway security.
Supply chain and runtime are linked: a verified server can still be misused by a prompt-injected agent. We cover that link in the MCP server supply chain is a runtime supply chain.
Permit MCP Gateway, which is generally available, is the runtime half of this picture. As shared in our MCP Gateway docs, it is a proxy between MCP clients such as Cursor, Claude Desktop, Claude Code and VS Code and the MCP servers they connect to. For each tool call it authenticates the human behind the agent, checks the call against policy in Permit, confirms the human consented to the agent's access level, and logs the decision. It classifies each tool by risk (read, write or destructive), denies by default until an admin grants access, and works with existing MCP servers without changing their code. You point the client at the gateway URL instead of the server URL.
Permit MCP Gateway is not a registry and doesn't replace one. Use a registry for discovery and publisher trust, and the gateway for per-call decisions.
If you want to try it, the MCP Gateway Free plan, labeled Community on the MCP Gateway pricing page, includes 10 tenants, 1 environment, 7 days of audit logs and up to 1,000 MAU across human and agentic identities.
What is the difference between an MCP registry and an MCP gateway? A registry lists MCP servers and verifies who published them. A gateway sits between clients and servers and decides whether each tool call may run, for which user and agent, and logs the result.
Does the official MCP Registry scan servers for security issues? No. Its documentation says security scanning is delegated to package registries and downstream aggregators, and its moderation policy says it doesn't remove servers with security vulnerabilities.
Can I use the official MCP Registry for internal servers? No. It lists only publicly accessible servers. Its docs recommend running your own private registry that implements the same OpenAPI spec.
Do I need a gateway if I only use verified servers? Verification tells you who published a server. It doesn't control what an agent does with it. Per-call authorization, consent and audit happen at runtime, which is the gateway's job.
To put per-call authorization, consent and audit in front of the MCP servers you already use, try Permit MCP Gateway.

Co-Founder / CEO at Permit.io