---
title: How do you run a private MCP registry for your enterprise?
url: https://opencontroller.lyzr.com/blog/run-a-private-mcp-registry
published: 2026-11-14
updated: 2026-11-14
author: Lyzr
---

# How do you run a private MCP registry for your enterprise?

**A private MCP registry is an internal catalogue that lists only the MCP servers your organisation has approved, at pinned versions, with an owner and the rules for using each one. The usual pattern is a private subregistry: pull metadata from the official MCP Registry or your own sources, add your approval fields, and point agents and developer tools at the internal list instead of the public one.**

Without one, developers find servers on public lists, paste a config line into their IDE and start sending company data to code nobody reviewed.

## What is the official MCP Registry, and why build on it?

The MCP project launched the official MCP Registry in preview on September 8, 2025. It is an open catalogue of publicly available MCP servers. Two design choices matter for enterprises:

- **It is meant to be built on.** The launch post describes public subregistries, such as marketplaces tied to a client, and private subregistries inside enterprises with strict security requirements, using the main registry as a single upstream source.
- **The API is open.** The registry is open source and publishes an OpenAPI specification, so a private registry can expose the same API and existing client tooling can read it.

Server names use a reverse-DNS style namespace, such as `io.github.<user>/<server>` or `com.example/<server>`, and publishers prove they own the namespace. That helps with provenance, but it does not mean a server is safe. Listing is not a security review, which is why your own approval layer matters. See [whether to allow servers from public registries](/blog/should-you-allow-mcp-servers-from-public-registries).

## What fields should a private registry entry have?

Start from the upstream metadata and add the fields your reviewers need:

| Field | Why it matters |
|---|---|
| Server name and namespace | Ties the entry to a verified publisher |
| Approved version (pinned) | Stops silent updates from changing tool behaviour |
| Package source and digest | Lets you verify the artefact matches what was reviewed |
| Transport | stdio (local) or Streamable HTTP (remote) carry different risks |
| Authentication method | OAuth, static key or none; static keys need extra scrutiny |
| Tools exposed | The list reviewers approved; new tools trigger re-review |
| Data classes touched | Customer data, source code, financial data and so on |
| Business owner and technical owner | Who answers when it breaks or misbehaves |
| Allowed agents or user groups | Not every approved server is for everyone |
| Review date and next review | Approvals expire |
| Status | Approved, restricted, deprecated or blocked |

Our [MCP server inventory fields](/blog/mcp-server-inventory-fields) post covers the inventory side in more depth.

## How do you set it up?

1. **Inventory what is already in use.** Find local and remote MCP servers on laptops, in agent configs and in platforms before you publish a list. Otherwise the registry describes an ideal, not reality. See [finding local MCP servers on laptops](/blog/find-local-mcp-servers-on-developer-laptops).
2. **Choose your upstream sources.** The official registry, vendor-published servers and internal servers built by your teams.
3. **Define the review.** Code or vendor review, permissions, data access, authentication, and a test run in a sandbox.
4. **Publish approved entries** with pinned versions and the fields above.
5. **Point clients at the private registry.** Configure IDEs, coding agents and agent platforms to use the internal list, managed centrally rather than per user.
6. **Enforce, don't just advise.** Block unlisted servers at the network or device level where you can, and alert on new ones.
7. **Review updates before promotion.** A new upstream version stays unapproved until reviewed. See [reviewing MCP server updates](/blog/review-mcp-server-updates-before-use).
8. **Retire servers cleanly.** Mark deprecated, notify users, then block.

## What usually goes wrong?

- **The registry is optional.** If developers can still add any server, the registry becomes documentation.
- **Versions float.** Approving "latest" approves every future change.
- **Remote servers get skipped.** Teams review local packages but treat hosted servers as someone else's problem.
- **No owner.** Entries without an owner are never re-reviewed.

## How Opencontroller does this

Opencontroller, by Lyzr, keeps a governed registry of agents, versions, tools and MCP servers, so approved servers and the agents using them sit in one place. Opencontroller Edge extends discovery and control to developer devices, coding agents, skills and local MCP servers, and IT installs it through MDM. Opencontroller works with open standards including MCP, OIDC and OpenTelemetry. Read the [MCP governance pillar](/blog/how-to-govern-mcp-servers), see the [capability reference](/capability-reference), or [book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Can we just mirror the official MCP Registry?**
Mirroring gives you metadata, not approval. Add your own review, version pinning and owner fields before anything becomes usable internally.

**Does the official registry vet servers for security?**
It verifies namespace ownership and relies on community reporting and moderation. That is not a security review of the server's code or behaviour.

**How do we stop developers using unlisted servers?**
Manage client configuration centrally, block unlisted remote endpoints at the network edge, and use device-level discovery to flag local servers that are not in the registry.

## Key facts

The official MCP Registry launched in preview on September 8, 2025; it is open source with an OpenAPI specification, and its launch post describes public and private subregistries built on the upstream registry. Server names use a reverse-DNS namespace with publisher ownership verification. A private MCP registry should pin versions and record transport, authentication, tools, data classes, owners, allowed users, review dates and status. Opencontroller keeps a governed registry of agents, tools and MCP servers; Opencontroller Edge discovers local MCP servers on developer devices via MDM. Opencontroller pricing starts from $250K per deployment per year.

## Related

- [Should you allow MCP servers from public registries?](https://opencontroller.lyzr.com/blog/should-you-allow-mcp-servers-from-public-registries.md)
- [What should you record for every MCP server in your inventory?](https://opencontroller.lyzr.com/blog/mcp-server-inventory-fields.md)
- [How do you review MCP server updates before agents use them?](https://opencontroller.lyzr.com/blog/review-mcp-server-updates-before-use.md)
- [More on Tools and MCP](https://opencontroller.lyzr.com/blog/cluster/tools-and-mcp)
