CyberCage – Security Platform for AI Tools and MCP Servers

TL;DR

TL;DR: CyberCage is an AI security platform that discovers, manages, and monitors MCP servers across a team — giving enterprises visibility into which tools are approved, what data is flowing to LLM providers, and who is running what.

Source and Accuracy Notes

⚠️ This section is MANDATORY. All links must be verified from actual source, not guessed.

• Project page: cybercage.io — verified via HTTP probe
• HN launch thread: news.ycombinator.com/item?id=46235072 — primary source for features and product description
• License: Not publicly disclosed as of publication

What Is CyberCage?

CyberCage is an enterprise-focused AI security platform built specifically for teams that use MCP (Model Context Protocol) servers in their AI workflows. The core problem it addresses: as teams adopt more AI tooling, there is little visibility into which MCP servers are running, what data they send to LLM providers, and whether they have been approved for use.

The platform positions itself as the control plane between your AI agents and the MCP servers they call.

“We built CyberCage because there wasn’t a good way to manage MCP servers across a team or see what’s actually flowing to LLM providers.” — CyberCage team, HN launch thread

Core Features

MCP Server Discovery and Approval

CyberCage supports both automatic and manual discovery of MCP servers. Once discovered, servers can be routed through an approval workflow before being made available to the team. This gives security and platform teams a gatekeeping mechanism — tools are not available to end users until they have been reviewed.

Management operates at two levels:

• Organization-wide — approve or block servers for the entire team
• Individual tool level — granular control over which specific tools within a server are allowed

Secure MCP Catalog

The platform integrates with GitHub’s MCP Catalog, allowing teams to pull in known, reviewed MCP servers from a curated source. This is intended to reduce the friction of discovering safe tooling while maintaining a review gate.

Operations and Tool Usage Tracking

CyberCage provides visibility into active MCP server usage across the team — which servers are running, how frequently specific tools are being called, and a high-level view of the AI tool landscape in a given organization. This helps answer questions like:

• Who is running what MCP servers?
• Which tools are actually approved?
• What data is being sent to external LLM providers?

Setup Overview

CyberCage is a hosted platform (no self-hosted option documented as of publication). Teams sign up and configure their MCP server inventory through the web interface at cybercage.io.

Typical onboarding steps (from product positioning):

1. Connect your team’s MCP server inventory
2. Run discovery to identify active servers
3. Review and approve servers via the approval workflow
4. Monitor active usage through the operations dashboard

Deeper Analysis

Who Is This For?

CyberCage targets enterprise teams that have adopted MCP-based AI tooling and need a governance layer around it. The approval workflow model is characteristic of enterprise SaaS — the platform assumes a separation between AI tool end users and a security/platform team that controls what those users can access.

For individual developers or small teams, the approval workflow overhead likely outweighs the benefit. The platform makes most sense at organizations where multiple teams are independently experimenting with AI agents and a central platform team needs to maintain visibility and control.

How It Compares

Traditional AI security tools focus on input/output filtering — scanning prompts and responses for sensitive data. CyberCage takes a different angle: it operates at the tool invocation layer, sitting between AI agents and the MCP tools they call. This gives it visibility into structured tool calls that might not be apparent from monitoring LLM API traffic alone.

The MCP Catalog integration with GitHub is notable — it signals an intent to align with the broader MCP ecosystem rather than building a walled garden of approved tools.

Practical Evaluation Checklist

• [ ] MCP server discovery finds all active servers in your environment
• [ ] Approval workflow supports both auto and manual discovery modes
• [ ] Tool-level (per-function) approval works correctly
• [ ] GitHub MCP Catalog integration surfaces curated tools
• [ ] Operations dashboard shows real-time server usage
• [ ] Data flow visibility correctly identifies data going to LLM providers
• [ ] Pricing is available before requiring sign-up

Security Notes

MCP server supply chain risk. MCP servers have access to the data that AI agents handle. A compromised or misconfigured MCP server can exfiltrate sensitive context passed to the LLM. CyberCage’s approval workflow is designed to mitigate this by requiring explicit approval before a server is available to users.

Data flow visibility. The platform monitors what data leaves your environment via MCP tool calls — useful for compliance and DLP teams that need to track PII or confidential business data moving to external LLM providers.

No published security audit or SOC 2 status was found on the product site as of publication. Enterprise buyers should request this during evaluation.

FAQ

Q: Is CyberCage open source? A: No open source offering or public source repository was found as of publication. CyberCage appears to be a closed-source SaaS product.

Q: Does CyberCage work with any MCP server or only those in GitHub’s MCP Catalog? A: Based on the HN launch description, CyberCage supports both GitHub MCP Catalog servers and custom/internal MCP servers via its discovery feature.

Q: What does the approval workflow look like? A: The product description mentions an approval workflow but specific details (number of approvers, integration with IdP like Okta/SAML, etc.) are not publicly documented. Prospective users should request a demo.

Q: Where does CyberCage send data for processing? A: This is not documented on the product site. Enterprise teams should clarify data handling during procurement.

Conclusion

CyberCage fills a specific niche: enterprise teams that have moved past AI experimentation and need operational control over MCP-based tool usage. The discovery and approval workflow model is straightforward, and the GitHub MCP Catalog integration is a pragmatic choice that reduces the friction of building a secure MCP tool inventory.

The main gap at publication is the lack of publicly available pricing, security certifications, or self-hosted deployment options — all of which enterprise security buyers typically require before engaging.

If your team is running MCP servers at scale without a central visibility layer, CyberCage is worth a demo. For smaller teams or individual developers, the enterprise governance model is likely overkill.