A single `forecast` tool your agent calls before any irreversible action (send money, run SQL, delete data, post content). Returns a risk score (0–100), a reversibility class, named red flags from 28 failure modes, and a gate: proceed / confirm / human-required.
Overview
BLACK_WALL gives your agent a brake. It exposes a single tool, forecast, that the agent calls before any irreversible action — sending an email, making a payment, running SQL, deleting data, posting content.
It returns:
- a risk score (0–100)
- a reversibility class + rollback cost (reversible / recoverable / irreversible)
- a gate: proceed / confirm / human-required
- named red flags from 28 documented failure modes (SQL_NO_WHERE, INTENT_MISMATCH, CROSS_ENVIRONMENT, MASS_RECIPIENT, PROMPT_INJECTION_LIKELY, …)
A normal email clears in one call; a DROP TABLE on prod comes back STOP / human-required — before it runs. The agent can't talk itself into the bad action, because the verdict is a hard gate.
Install
Works in Claude Desktop, Claude Code, Cursor, and Windsurf. Add one block:
{
"mcpServers": {
"blackwall": {
"command": "npx",
"args": ["-y", "blackwall-mcp"],
"env": { "BLACKWALL_API_KEY": "bw_live_..." }
}
}
}
Setup: get a free API key at https://blackwalltier.com (~100 forecasts/month, no card). A verdict takes ~4–8s.
A BLUETIER product.
Server Config
{
"mcpServers": {
"blackwall": {
"command": "npx",
"args": [
"-y",
"blackwall-mcp"
],
"env": {
"BLACKWALL_API_KEY": "bw_live_your_key_here"
}
}
}
}