Agenvoid

A MCP & Skill security manager for AI Agents like Claude Code, OpenClaw, Codex CLI. Monitor, audit and control what your AI agents can access before they act. Add an MCP once, distribute to your agents with time-based access, per-project control, and zero keys exposed. For Mac & Linux.

The Problem

Every local AI agent, whether it's Claude Code, OpenClaw or Codex, needs MCPs and Skills to be useful. But right now, there's no control layer between them and your system.

Scattered ConfigYour MCP servers and API keys live in plain-text JSON files across every agent's config folder.

No BoundariesOnce a Skill or MCP is granted, your agent has permanent, unscoped access. No time limits, no project boundaries.

No VisibilityYou have no log of which MCPs your agents are calling, what data they're reading, or if their config has been tampered with.

How Agenvoid Works

Add your MCPs and Skills into Agenvoid once. Then assign them to agents, scoped by project, limited by time, secured by your OS Keychain. There's a UI for convenience and CLI for manual control or automations.

Agenvoid sits between your local agents and the tools they use. Instead of each agent holding its own keys and config, Agenvoid becomes the single source of truth. You decide which agent gets access to which MCP or Skill, for how long, and in which project context.

When a grant expires, access is revoked automatically. No cleanup, no forgotten permissions, no stale keys sitting in a dotfile somewhere.

Built for Control

CENTRAL HUB

One Place for All MCPs and Skills

Register your MCP servers and Skills once in Agenvoid. Distribute them to Claude Code, OpenClaw, Codex or any local agent without duplicating config files.

GRANTS

Per-Agent, Per-Project, Time-Based

Give Agent A access to your GitHub MCP for Project X, for 1 hour. Give Agent B read-only Slack access for the day. Every grant is scoped and auto-revokes.

KEYCHAIN

Keys Locked in OS Keychain

API keys and auth credentials are stored in the macOS Keychain or Linux secret store. Never in config files, never visible to agents. Agenvoid proxies authenticated requests so agents never see the secrets.

AUDIT

Full Request Log and Tamper Monitoring

Every MCP call and Skill invocation is logged in a real-time stream. Agenvoid also monitors agent config files for unauthorized changes so you'll know if something was added behind your back.

Join the Private Alpha (Limited Seats)

We're onboarding a small group of Claude Code and OpenClaw power users to stress-test Agenvoid's security layer before the public V1 launch. Limited slots remaining for the first alpha test phase.

Frequently Asked Questions

What is Agenvoid?

Agenvoid is a local security proxy for AI agents. It sits between your agents (Claude Code, OpenClaw, Codex) and the MCPs and Skills they use, giving you granular control over what each agent can access.

How is this different from just editing my agent config files?

Config files give all-or-nothing access with no expiry. Agenvoid lets you scope access per agent, per project, with automatic time-based revocation. Your keys never touch a config file.

Where are my API keys stored?

On macOS they go into the system Keychain. On Linux they are stored in the native secret store (libsecret / GNOME Keyring). Agents never see the raw credentials. Agenvoid proxies authenticated requests on their behalf.

What does tamper monitoring do?

Agenvoid watches your agent config files for unauthorized changes. If an MCP or Skill is added outside of Agenvoid, you get an alert. This prevents agents or scripts from silently expanding their own access.

Which agents are supported?

Any local agent that uses MCP servers or Skills. Claude Code, OpenClaw, Codex and similar tools. If it reads from a local config file to connect to MCPs, Agenvoid can manage it.

Is Agenvoid free?

Once Agenvoid is released it will cost $79 per year for individuals and $179 per year for business users (ex taxes). We will offer a introduction price of $39 at launch of beta. Join the wait list to keep updated.