A few months ago we rolled out ToolHive as a lab project, with the intent of making MCP servers simpler and more secure to deploy and manage. Since then, the MCP ecosystem has been humming, the ToolHive project has been picking up steam, and a small community of practitioners has been forming.
In short, things have been getting real, and today is yet another proof point, as we’ve released a slate of new ToolHive capabilities.
What ToolHive is all About
We believe that the ultimate beneficiaries of MCP will be knowledge workers. Those workers will want/expect AI agents and models to complete real, important work; and the only way to do that is to give them controlled access to enterprise data, systems and services. That requires a protocol like MCP.
Our immediate focus with ToolHive is creating value for enterprise developers. At service providers, they’re responsible for packaging products as MCP servers that can offer new and better customer experiences. And at enterprises, those developers are accountable for setting up and managing a growing estate of MCP servers. In both cases, those enterprise developers will need (1) simple, modular integrations and (2) fine-grained controls (especially of security to appease AI governance committees).
And that brings us back to …
Today's ToolHive Release
The latest capabilities added to ToolHive are designed to simplify the discovery and setup of MCP servers. Specifically:
Point-and-Click User Interface
Up to this point, ToolHive has only been available for use via a command-line interface. Today’s release introduces a clean UI that makes it easy for any enterprise developer to find and run pre-packaged MCP servers, or configure and deploy any custom server.
Pre-vetted MCP Server Registry
One of the challenges with early MCP servers is that they can be faulty or flawed. For example, one of the most commonly used MCP servers is Fetch, which retrieves and processes web content for an AI agent or model. Where it makes sense, we are rewriting and hardening key MCP servers. Our ‘GoFetch’ MCP server is a great example.
Of course, MCP servers can be more than faulty; there are examples of MCP server poisoning, MCP server typosquatting, prompt injections and more. Since the provenance of an MCP server is rarely known, it’s hard to know which servers to trust.
ToolHive includes a registry of pre-vetted MCP servers. We’re not trying to house hundreds or thousands of MCP servers; our focus is on a highly curated set of the most useful MCP servers, verified and/or improved upon by our security-minded team.
Client Integrations
Part of keeping MCP simple is tucking it into an enterprise developer’s existing workflow; making it minimally disruptive. And so, ToolHive handles the configuration of servers for popular tools like Copilot in Visual Studio Code, Cursor, Claude Code and more. Instead of detouring through multi-step configuration processes, ToolHive just takes care of it.
Consistent Runtimes
More than 5,000 MCP servers have been published in 2025 alone. All those servers are developed using many different languages, and are written to different standards. This variety is untenable for an enterprise developer; but it’s also a problem we feel has been solved before … with containers. So, with a single click or command, ToolHive containerizes an MCP server. This introduces far more consistency and control, plus all the other benefits you’ve come to expect from containers (portability, isolation, etc.)
Now, there’s plenty more we can and will do to further simplify use of MCP, and this is a great time to remind you that ToolHive is open source, so you can get involved in shaping our roadmap and contributing via our GitHub repo or engaging with us on Discord.
But for the moment, let’s shift our focus to how ToolHive is better securing MCP servers. After all, if enterprise developers are going to drive real adoption of MCP, the right guardrails will need to be in place. We’ve started with:
Network Isolation
MCP is still maturing and the community is constantly discovering … quirks. Some of those quirks can justifiably terrify an enterprise developer. For example, MCP servers can make network calls in the background, without your knowledge. That means they could expose you / your enterprise to data exfiltration and malicious activity.
That’s why it was imperative for us that ToolHive include network isolation. As part of this release, ToolHive now blocks unauthorized connections and provides a full audit trail should you ever need it.
Built-in Security
We’re also working to make sure security is increasingly ‘built-in’ at every turn. As a first example, many client implementations require developers to store API tokens in plain text configuration files. That strikes us as a bad idea. So, with ToolHive, we’ve included secure secrets management via an encrypted vault and/or 1Password integration.
We’ve also included an OAuth-based framework for authentication and authorization, enabling integrations with your identity provider of choice. You can read more about those decisions on our dev.to profile, including this recent post on access levels and auditability.
Next Steps
We’re busy over here plowing through an ambitious roadmap and digging in with design partners. There are lots of ways you can get involved.
First, if you’re an enterprise developer who is tasked with helping your organization use MCP in production, we’d like to talk with you about design partnership. You should reach out now to hello@stacklok.com.
Second, if you’re keen to stay involved with this fast-moving space and the ToolHive project, you should check out and follow our GitHub repo and engage with our friendly team in Discord.
And third (and perhaps most important), since ToolHive is free and open source, you can head over to toolhive.dev right now to find the download link that’s right for you. We encourage you to dig into our docs (we’ve put a lot of care into them), including a quickstart that will help you deploy your first, pre-vetted MCP server.
We’re excited about what we’ve built, but even more so about what’s ahead.
Top comments (0)