Introducing Overwatch: Code Agent Security

Code Agent Security

When developers open their IDEs today, they’re not just writing code. They’re working alongside agents, tools, and servers that can generate, analyze, and even ship code on their behalf. The rise of the Model Context Protocol (MCP) has made it easier than ever for these agents to plug directly into local environments. But the line between helpful and harmful servers is far thinner than most people realize.

We’ve seen what happens when this blind spot goes unchecked. An unauthorized MCP server quietly configured in the background can siphon off source code. A malicious server masquerading as a trusted tool can inject instructions that compromise security from the inside. And the worst part? These risks rarely show up in the places security teams are watching. By the time traffic hits a firewall or runtime monitor, the damage may already be done.

Why We Built Overwatch

Overwatch is a lightweight plugin that lives inside the IDE, right where the work happens. Its role is simple but vital: keep watch over local MCP configurations, spot unrecognized or malicious servers, and shut them down before they cause harm. Think of it as an extra set of eyes—always scanning, checking against trusted sources, and raising the alarm when something doesn’t look right.

Each time you open your IDE, Overwatch runs an automatic scan of active MCP connections. That scan is tuned for the kinds of vulnerabilities that matter in practice: tool poisoning attacks that bend agents to an attacker’s will, indirect prompt injections that sneak in through side channels, jailbreak attempts that strip away safety guardrails, and the quiet exfiltration of secrets or personally identifiable information. These threats live in the gray space of agent development, and Overwatch is designed to surface them before they take hold.

For organizations, the challenge isn’t just identifying what’s dangerous—it’s also making sure only what’s approved can run. Overwatch introduces enterprise controls that let teams define which MCP servers are trusted and which should be blocked outright. By tying into existing identity and access management roles, white lists and black lists flow directly down to developer machines. That means a developer writing code locally gets the same protection as one deploying to production: consistent guardrails without friction.

And Overwatch doesn’t stop at the obvious servers. It also watches for the unknowns—local MCPs spun up on a laptop, endpoints that appear without authorization, or tools introduced into a project without review. Instead of letting those connections operate in the shadows, Overwatch makes them visible, giving both developers and security teams the context and control they need.

The Developer Experience

We knew from the start that security had to fit naturally into the way developers work. Nobody wants a clunky tool that slows them down or interrupts their flow. Overwatch was built to be invisible until it’s needed, keeping developers confident that the tools they rely on every day are safe.

One of the things we’re proudest of is what you don’t notice. Overwatch is written entirely in Rust, which means it runs close to the metal, safe by design, and incredibly fast. The result is a plugin that consistently uses less than twenty megabytes of RAM—lighter than a browser tab, smaller than most extensions already running in your IDE. In practice, it means Overwatch can keep a constant eye on local MCP configurations without stealing cycles from your editor or slowing you down. Security doesn’t have to come at the cost of performance, and with Overwatch it doesn’t.

For security teams, Overwatch finally shines a light on part of the development process that has long been invisible. MCP connections no longer happen in the dark. They’re visible, auditable, and governed from the moment they’re spun up locally. For organizations, this closes a gap attackers are already probing—right at the intersection of developer productivity and enterprise security.

A Secure Agentic Future

Overwatch is part of a larger vision at Javelin: securing the agent-powered workplace from the ground up. Alongside Rampart and our broader MCP security work, it ensures that whether agents are running in production, in pipelines, or on laptops, they’re operating in an environment where trust is earned, not assumed.

We’re excited to bring Overwatch to the community, starting with support for popular IDEs like VS Code and JetBrains. Installation takes just a few minutes, and from that moment on, your IDE has a built-in safeguard against unauthorized and malicious MCP servers. The future of coding is fast, collaborative, and agent-driven. With Overwatch, we can make sure it’s secure too.

If your developers are already using code agents like Cursor, Copilot, Windsurf or Claude Code, now’s the time to make sure they’re secure. Reach out—we’d love to show you how Overwatch can help.

Request a demo