Your developers run Claude Code, Cursor, Copilot, Windsurf, and a growing pile of MCP servers. Each one can read files, run shell commands, and hold API keys. This audit answers one question in writing: what can they actually touch, and what should you lock down first.
Up to 10 developer machines. Every agent, every config, every credential in scope. Delivered in 5 business days.
Every AI coding tool on every machine, with version, config locations, and whether it runs with a permission prompt or auto approves everything.
Each MCP server your agents load: what package it runs, who publishes it, whether it has shell or network access, and what environment it inherits.
API keys sitting in mcp.json, tokens in shell profiles that every agent inherits, .env files inside agent scope. Found, rated, and mapped to who can read them.
Which agents can reach your home directory, SSH keys, cloud credentials, and other projects. Most teams find at least one agent with everything.
Allow lists, deny lists, sandbox settings, and package install rights, per agent. We flag the configs that let a prompt injection run anything it wants.
A written report with a risk score per machine and a remediation list ordered by impact. Plus a 60 minute walkthrough so your team knows what to do Monday.
No agents installed on your side. No access to your machines required.
Each developer runs npx snuf --format json --output snuf.json. It reads configs, never modifies them, and makes zero network calls. You review the output before sending anything.
We go over your team, the tools in use, what you're most worried about, and what compliance questions you're being asked. Free, no commitment.
Manual review of every scan result and every config, plus checks the scanner can't do on its own: MCP package provenance, rules files, CI agent tokens, shared secrets.
Written report, risk score per machine, fix list ordered by impact, and a 60 minute call to go through it with whoever owns the fixes.
Most teams handle the fix list themselves in a day or two. If you'd rather I do it, that's a separate fixed quote after the audit.
These show up on most teams. None of them are exotic. All of them are one prompt injection away from an incident.
OpenAI, Anthropic, GitHub, and database credentials pasted straight into mcp.json, readable by every agent and every MCP server on the machine.
An agent started from the home directory can read SSH keys, AWS credentials, browser profiles, and every other project on the disk.
Permission prompts turned off for speed. Fine until a README, an issue, or a web page tells the agent to run something.
Servers installed from a blog post, running arbitrary npm or pip packages with shell access and the developer's full environment.
CLAUDE.md, .cursorrules, and AGENTS.md files that instruct agents to run commands, fetch URLs, or write outside the project.
Cloud and API tokens exported in shell profiles, silently inherited by every agent and every subprocess they spawn.
Every one of these hit developer machines through the same three things: an MCP config, a rules file, or a permission setting nobody reviewed.
Malware invoked Claude Code, Gemini CLI, and Amazon Q with permission bypass flags to hunt for credentials. Over 2,300 secrets stolen from developer laptops.
A prompt injection rewrote a project's MCP config. Once approved, the config was silently swapped for one that ran attacker commands on every open.
A hook in a repo's settings file executed before the trust dialog appeared. A second flaw let a project file redirect the API key to an attacker's server.
GitGuardian found 24,008 credentials committed inside MCP config files on public GitHub. Over 2,100 were still valid when tested.
73 Microsoft repositories were seeded with a Claude Code session hook and MCP configs. Opening the repo in Claude Code, Cursor, or Gemini CLI ran a credential harvester.
Approval dialogs in six coding assistants showed a harmless file name that was a symlink to SSH authorized keys. The user approved what they could not see.
Fixed price. Written report in 5 business days. Start with the free scan today.
Or email directly: [email protected]