All posts
AI
November 13, 2026

What Is an MCP Registry?

What Is an MCP Registry?

See how Airia governs every MCP connection. Explore Airia MCP Gateway

A developer adds an MCP server to their coding assistant in about thirty seconds. A product team connects an agent to a CRM server they found on GitHub. Someone in finance installs a spreadsheet tool from a community list. Each choice is reasonable. Together, they leave the organization with no shared answer to a simple question: which MCP servers are our agents allowed to use?

An MCP registry is how you answer that question. This guide explains what an MCP registry is, how public and enterprise registries differ, the security risks a registry should address, and what to look for when you choose one.

What is an MCP registry?

An MCP registry is a catalog of Model Context Protocol (MCP) servers that AI clients and agents can search, evaluate, and connect to. It stores metadata about each server, such as its name, publisher, version, where it runs, and how to connect, rather than the server’s code.

MCP is the open standard that lets AI applications call external tools and data sources through a common interface. An MCP server exposes tools, such as “create ticket” or “query database,” and an MCP client, such as an agent or AI assistant, calls them. The registry sits one step earlier. It tells clients which servers exist and how to reach them.

Without a registry, every developer configures servers by hand, every agent has its own list, and nobody has a complete picture of what’s connected.

How an MCP registry works

Most registries follow the same basic flow.

  1. Publish. A server maintainer, or an internal platform team, adds a server entry with its metadata and connection details.
  2. Discover. Clients query the registry to find servers by name, category, or capability.
  3. Connect. The client uses the registry entry to configure the connection, often with credentials supplied separately.
  4. Maintain. Entries are updated as servers change versions, and removed when they’re retired or found to be unsafe.

The key design choice is that a registry describes servers but doesn’t run them. The official MCP Registry charter is explicit on this point: the registry is “a metadata catalog, not a package registry.”

Public MCP registries vs. enterprise MCP registries

The phrase “MCP registry” covers two different things, and the difference matters for anyone responsible for security or AI governance.

The official MCP Registry

The MCP project launched the official MCP Registry in preview on September 8, 2025 as “an open catalog and API for publicly available MCP servers.” It gives server maintainers one place to publish and gives client builders one source to pull from. Community members can flag servers that break moderation guidelines, and maintainers can remove them.

The official registry was designed to feed other registries. Its API specification is open, so public marketplaces and private enterprise registries can build on the same format. The project also states its limits clearly. The preview came with no data durability guarantees, and the working group lists an enterprise-ready registry implementation as out of scope.

Enterprise and private MCP registries

An enterprise MCP registry answers a different question. A public registry asks, “What servers exist?” An enterprise registry asks, “What servers are we allowed to use, who can use them, and which tools inside them?”

Public MCP registry Enterprise MCP registry
Purpose Help anyone discover available servers Control which servers your organization uses
Who adds entries Server maintainers Your platform, security, or AI team
Access Open to everyone Scoped by role, team, or user
Trust model Namespace ownership and community moderation Internal review and approval before use
Permissions Not covered Set per server, per tool, and sometimes per parameter
Audit Not covered Records of which agent called which tool, and the result

Most enterprises need both. The public registry is a good source of candidates. The enterprise registry is where those candidates get reviewed, approved, and governed.

Why MCP servers need governance, not just a list

MCP servers act on real systems. A tool that can read a CRM can also export it, and a tool that can open a ticket can often close one. That makes the servers themselves part of your attack surface.

The OWASP MCP Security Cheat Sheet describes several risks a registry alone does not solve:

  • Tool poisoning. Hidden instructions in a tool’s description or schema steer the model toward actions the user never intended.
  • Rug pulls. A server changes its tool definitions after someone approved it, so a trusted tool becomes a harmful one.
  • Tool shadowing. A malicious server’s tool description changes how an agent uses tools from other, trusted servers.
  • Supply chain attacks. Teams install unreviewed or compromised server packages from public sources.
  • Excessive permissions. Servers request broad access that turns one mistake into a large exposure.

A list of approved servers is the starting point. Governance comes from what happens when an agent actually tries to use one.

MCP registry vs. MCP gateway

A registry and a gateway are related but different.

  • The registry is the catalog. It defines which servers exist and which are approved.
  • The gateway is the checkpoint. It sits between agents and servers at runtime, applies policy to every tool call, and records what happened.

A registry without a gateway tells agents where to go but can’t stop them from doing the wrong thing when they get there. A gateway without a registry has nothing consistent to enforce against. Enterprise teams get the most value when both share the same identities, approvals, and policies.

What to look for in an enterprise MCP registry

Use these questions to evaluate any registry, whether you build it or buy it.

  1. Is there an approval step? New servers, including custom ones your teams build, should be reviewed before any agent can use them.
  2. Can you control access below the server level? Server-level access is too coarse. Look for permissions by tool, and ideally by parameter, assigned by role.
  3. Are tool definitions scanned? Tool names and descriptions should be checked for hidden prompt injection before they reach a model.
  4. Can you tell safe tools from risky ones? Labels such as read-only or destructive help admins decide what each role should see.
  5. Is every call logged? You need a searchable record of each tool call, including calls that were denied or failed.
  6. Does it scale without flooding the context window? Large catalogs need a way for agents to find the right tool without loading every definition on every call.
  7. Does it fit how people actually work? One governed endpoint that works across clients is easier to adopt than per-device setup.

How Airia approaches the MCP registry

Airia treats the registry and the gateway as one governed layer. Through Airia’s MCP Gateway, admins choose which servers from a catalog of pre-configured MCP servers, plus any custom servers, can be used at all. No one can add a server to a gateway or agent until it’s approved.

From there, admins assign access at the app, tool, and parameter level by role. A single Dynamic MCP Gateway address serves every user with a tool set matched to their role, so there’s no per-device configuration. Every tool’s name and description is scanned for hidden prompt injection before it’s exposed, and tools carry annotations so admins can filter for read-only or destructive actions.

Every tool call, whether it succeeded, was denied, or errored, is logged and searchable. For large catalogs, Airia Radar lets agents search for the tools they need instead of loading every definition into the context window.

Because the MCP layer runs on the same platform as Airia’s AI guardrails, agent constraints, and governance dashboard, the policies that govern your models also govern the tools your agents call.

Conclusion

An MCP registry answers which servers exist. An enterprise MCP registry answers which servers, tools, and users are allowed, and a gateway makes sure that answer holds at runtime. Start with the public registry as a source of candidates, approve what your organization needs, set permissions at the tool level, and log every call. That’s how MCP moves from developer convenience to governed infrastructure.

Find out which MCP servers your agents can reach today. See how Airia approves servers, scopes tools by role, and logs every call in one governed gateway. Request a demo

Put these ideas to work.

Schedule a 30-minute walkthrough with our team.

Talk through your use case