# red-harness MCP server

> Provides an evidence-bound reverse-engineering harness that verifies AI agent claims against static analysis and sandboxed execution.

- Repository: https://github.com/LiebertDev/red-harness
- Page: https://hostedmcp.org/servers/liebertdev/red-harness
- Transport: —
- Stars: 0
- License: Apache-2.0

## Environment variables

| Name | Required | Secret | Description |
|---|---|---|---|
| `RH_DATA_ROOT` | no | no |  |
| `RH_GHIDRA_DIR` | no | no |  |

## FAQ

### How does Red Harness ensure AI agent claims are accurate?

It uses a deterministic verifier that re-checks every claim against an evidence ledger containing IDs, quotes, and addresses, rejecting unsupported claims.

### Can the server execute the binaries it analyzes?

Yes, it can perform controlled execution of the target within a Windows Sandbox when the scope manifest allows it.

### What analysis engines are supported?

The server supports both Ghidra and IDA Pro as engines for static analysis tools.

### Does the server require an API key?

No, the core installation does not require a model or an API key to function.