A New Tool Called Traceforce Helps Companies See What AI Apps Their Employees Are Actually Using
Traceforce, co-founded by Xia and Varun, has launched with a security platform that helps organizations see and control AI applications running on company devices, including laptops and virtual machines. The product is currently deployed across more than 1,000 devices at 10 organizations, according to a launch post on Hacker News dated July 16, 2026. Hacker News
The core problem Traceforce addresses is one that many security teams have not yet fully mapped: which AI apps employees are using, what data those apps can reach, and what risks those connections introduce. Traceforce discovers which AI applications are in use and how they are connected to other data sources via MCPs (Model Context Protocols). An MCP is essentially a bridge that lets an AI application connect to things like databases or internal systems. On average, the company reports finding over 15 AI applications per device, with each application connected to 5 to 10 MCPs. Hacker News
Installation is lightweight: a small program and a browser extension on each device. Setup takes approximately 15 minutes, per the company's website. Traceforce.ai Within 30 minutes of installation, Traceforce begins uploading live data to a central dashboard that displays all AI agents and apps running across the organization's devices. Hacker News
The platform goes beyond monitoring. It provides real-time threat blocking and automated remediation, and it monitors command-line tools in addition to browser-based AI apps and MCP connections, according to the company's website. Traceforce.ai In practice, Traceforce has already helped customers find exposed passwords in MCP configurations, prevent API keys from leaking through AI-generated code, and warn developers before executing potentially destructive commands such as DROP TABLE. Hacker News
Privacy architecture is a notable design decision. By default, Traceforce collects only basic information about AI applications, MCPs, and tools running on a device, not the content itself. All content inspection happens locally, on the device. User prompts are never stored unless an organization's security administrators explicitly turn that capability on. Hacker News
The product also ships an open-source tool called mcp-xray, hosted on GitHub. GitHub The tool allows security teams to test MCP configurations for security weaknesses on their own, without needing the commercial platform. Hacker News
Before building Traceforce, the founders spoke with over 50 Chief Information Security Officers and Chief Information Officers to validate the problem space. Co-founder Xia was previously Director of Engineering at Clumio, which was acquired by Commvault in October 2024. Hacker News
Traceforce targets small to medium-sized companies with 200 or more employees that are rapidly adopting AI coding assistants, ChatGPT, Claude, and MCPs. The company offers a free trial via its website and has published a video demo on YouTube. Hacker News
The product's enforcement model is what the founders call a "warn and acknowledge" approach: developers retain the freedom to work as they choose, but receive warnings before taking actions that could result in costly mistakes. Hacker News
The broader context here is worth pausing on. Fifteen-plus AI applications per device, each wired into 5 to 10 MCPs, means a single laptop may expose dozens of data-source connections that no security team has inventoried. That is a dramatically larger security risk than the unchecked spread of cloud apps that worried IT departments a few years ago, and it has emerged quickly. MCP adoption has accelerated through 2025 and into 2026, and the protocol's open design means that a poorly configured MCP can expose databases, internal systems, or file systems to any AI agent that connects to it.
The decision to handle content inspection locally rather than in the cloud is a pragmatic one. It sidesteps the thorniest privacy objection: the idea that a security tool would need to see or store employee prompts to be effective. Whether basic metadata alone proves sufficient for detection at scale is a question that broader deployment will answer.
The open-sourcing of mcp-xray is also a telling signal. Testing tools for MCP configurations are not widely available, and releasing one freely gives the broader security community a concrete way to assess MCP risk without committing to a commercial product. It also positions Traceforce as a contributor to the emerging MCP security ecosystem rather than solely a vendor within it.
For organizations now managing the intersection of AI coding assistants, automated AI workflows, and MCP-connected data sources, Traceforce is betting that monitoring at the device level is the right place to provide visibility and control. The early traction, 1,000 devices across 10 organizations, suggests the bet is resonating with at least a segment of the market. Whether that scales beyond smaller company deployments into larger enterprises with more complex device-management needs will depend on factors the company has not yet had to address at this stage.
