# Revoke Probe

> Local-first reachability check for AI coding-agent and MCP workflows

- **Type:** MCP server
- **Install:** `agentstack add mcp-revoke-trust-network-revoke-probe`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [Revoke-Trust-Network](https://agentstack.voostack.com/s/revoke-trust-network)
- **Installs:** 0
- **Category:** [Security](https://agentstack.voostack.com/c/security)
- **Latest version:** 0.1.0
- **License:** MIT
- **Upstream author:** [Revoke-Trust-Network](https://github.com/Revoke-Trust-Network)
- **Source:** https://github.com/Revoke-Trust-Network/revoke-probe
- **Website:** https://revokenode.io/probe

## Install

```sh
agentstack add mcp-revoke-trust-network-revoke-probe
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# Revoke Probe

Local-first reachability check for AI coding-agent and MCP workflows.

[](https://opensource.org/licenses/MIT)
[]()
[]()

Revoke Probe helps you answer one narrow question:

> What sensitive-looking local surfaces could this developer workflow reach?

It scans for common credential-adjacent path patterns and can be paired with
**fake canaries** to produce a **sanitized report** you can share after review.

No telemetry. No source upload. No real secret values in the report. No
vendor-blame claims.

`revoke-probe` is the free local probe used by Agent Leak Bench.

## Install

### macOS

```sh
brew install Revoke-Trust-Network/tap/revoke-probe
```

### Linux / WSL / Git Bash

```sh
curl -fsSL https://raw.githubusercontent.com/Revoke-Trust-Network/revoke-probe/v0.1.0/install.sh -o revoke-probe-install.sh
sh revoke-probe-install.sh
```

### From source

```sh
git clone https://github.com/Revoke-Trust-Network/revoke-probe
cd revoke-probe
./revoke-probe --scan
```

## Usage

```sh
$ revoke-probe --scan
```

Output is a structured report of which AI-tool-accessible paths match common
high-value patterns, sorted by risk tier. The report is designed to be safe to
share after review because it must not include real secret values.

For automation:

```sh
revoke-probe --scan --json
revoke-probe --scan --path /path/to/test/root
```

## What it scans (path pattern catalog)

See [docs/path-patterns.md](docs/path-patterns.md) for the full list.
TL;DR — 18 path families across SSH/cloud creds, dotenv variants, browser
profiles, and framework config dirs.

## Fake canaries

The benchmark uses fake values only:

- fake AWS key marker
- fake Stripe key marker
- fake GitHub token marker
- fake OpenAI key marker
- fake SSH private key marker

Do not paste real credentials into issues, reports, screenshots, posts, or
support emails.

## What it proves

- Which sensitive path families are reachable on this machine.
- Which findings are worth reviewing before a team standardizes an AI coding workflow.

## What it does NOT prove

- It does not prove a vendor stole your data.
- It does not upload anything anywhere.
- It does not require real secret values.
- It does not prevent agents from reading these files by itself.
- It does not replace endpoint security, MDM, or secrets management.
- It does not modify your filesystem outside controlled benchmark fixtures.

## Why this exists

AI coding tools moved fast. Team policy did not. Most developers can name the AI
assistant they installed, but they cannot quickly answer a more operational
question:

> What can my current AI coding stack touch on this machine?

This benchmark exists to make that question visible, reproducible, and boring
enough to discuss without vendor-blame theater.

## Contributing

PRs welcome, especially:

- Additional dotfile patterns we missed
- Translations of the report output

## Roadmap

- [x] Path pattern catalog
- [x] Shell CLI scanner
- [x] JSON output mode
- [x] Homebrew tap
- [ ] eBPF-backed real-time read tracing (Linux)
- [ ] CI integration mode

## Pro tier — `revoke-node`

`revoke-probe` focuses on scan + report.

`revoke-node` is the commercial product line around the same problem space.
This README intentionally avoids capability promises that are not verified
inside this repo.

## Team Pilot

Teams using Cursor, Claude Code, Codex CLI, Windsurf, VS Code extensions, or MCP
servers can request a 30-day audit:

```text
support@revokenode.io
Subject: Team Pilot
```

## License

MIT © 2026 Revoke-Trust-Network

## Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [Revoke-Trust-Network](https://github.com/Revoke-Trust-Network)
- **Source:** [Revoke-Trust-Network/revoke-probe](https://github.com/Revoke-Trust-Network/revoke-probe)
- **License:** MIT
- **Homepage:** https://revokenode.io/probe

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** yes
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** yes
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/mcp-revoke-trust-network-revoke-probe
- Seller: https://agentstack.voostack.com/s/revoke-trust-network
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.