Traceforce, a YC S26 startup, launches a platform that provides visibility and control over AI applications like ChatGPT and Claude across all company devices. It monitors AI app usage and MCP connections, and includes an open-source pentesting tool for MCP vulnerabilities. The founders aim to help security teams detect and prevent unsafe AI actions in real time.
SynthePulse Insight · AI deep reading
Version 2 · 1 source
Traceforce provides enterprises with visibility and control over AI applications through lightweight agents and browser extensions, while balancing security and user privacy through local content inspection and privacy-preserving design.
Traceforce is an enterprise AI application security monitoring platform designed to provide visibility and control over AI applications such as ChatGPT and Claude. It installs a lightweight binary and browser extension on each device, uploading device data to the company dashboard within 30 minutes, showing real-time activity of all AI agents and applications.
The platform not only monitors the application layer but also MCP servers and their supply chain. Founder Xia, in response to comments, stated that they run a hosted version of mcp-xray in the backend, continuously performing penetration testing on MCPs.
Traceforce emphasizes transparency: by default, only metadata and telemetry are collected, content inspection is performed locally on the device, and user prompts are not stored unless explicitly configured by the organization's security administrator. This design aims to balance security monitoring with user privacy.
The founder mentioned that once end users understand what is monitored and shared, they feel reassured by the on-device protection layer, which helps reduce concerns about data leakage.
Traceforce is currently deployed on over 1,000 devices across 10 organizations, discovering an average of more than 15 AI applications per device, each connecting to 5-10 MCPs. Founder Xia has spoken with over 50 CISOs and CIOs, believing this is a pressing need across industries.
In Hacker News discussions, some commenters pointed out intense market competition, including emerging companies like Runlayer and Bluerock. Xia responded that Runlayer might suit scenarios with known requirements, but customer feedback indicates they lack visibility into the AI applications and MCPs employees actually use. Xia acknowledged the crowded market but argued it proves the problem is real, emphasizing their differentiation through community-first and open-source-first approaches.
Traceforce's binary is built with Go, and the browser extension is written in Node.js. The most difficult part is building a complete connection graph between AI applications, MCPs, and tools, and identifying vulnerabilities and attack paths introduced by these connections. Traditional security tools like EDR and CASB cannot see application-level activity inside AI applications.
Traceforce open-sourced mcp-xray, a dynamic MCP penetration testing tool used to detect vulnerable MCPs. This reflects its open-source-first strategy.
Traceforce helps customers identify exposed plaintext keys in MCP configurations, prevent API key leakage through AI-generated code, and warn developers before they execute potentially destructive commands like "DROP TABLE". Their "warn and confirm" approach has been well-received, giving developers freedom while avoiding costly mistakes.
The founder stated that they work closely with end users of the product, and users feel comfortable once they fully understand what is monitored and shared, as they have a strong layer of protection.
This report is based on Traceforce's launch post on Hacker News and comments, which are self-promotional; deployment data and use cases have not been independently verified. Competitive information in comments comes from third parties and may be subjective.
Traceforce provides enterprises with visibility and control over AI applications through lightweight agents and browser extensions, while balancing security and user privacy through local content inspection and privacy-preserving design.
Primary report
Primary source