Blackwall MCP server
{
"mcpServers": {
"blackwall": {
"command": "npx",
"args": [
"-y",
"blackwall-mcp@latest",
"blackwall-mcp"
],
"env": {
"BLACKWALL_API_KEY": "<blackwall_api_key>",
"BLACKWALL_BASE_URL": "https://..."
}
}
}
}{
"mcpServers": {
"blackwall": {
"command": "npx",
"args": [
"-y",
"blackwall-mcp@latest",
"blackwall-mcp"
],
"env": {
"BLACKWALL_API_KEY": "<blackwall_api_key>",
"BLACKWALL_BASE_URL": "https://..."
}
}
}
}{
"mcpServers": {
"blackwall": {
"command": "npx",
"args": [
"-y",
"blackwall-mcp@latest",
"blackwall-mcp"
],
"env": {
"BLACKWALL_API_KEY": "<blackwall_api_key>",
"BLACKWALL_BASE_URL": "https://..."
}
}
}
}{
"servers": {
"blackwall": {
"command": "npx",
"args": [
"-y",
"blackwall-mcp@latest",
"blackwall-mcp"
],
"env": {
"BLACKWALL_API_KEY": "${input:blackwall_api_key}",
"BLACKWALL_BASE_URL": "${input:blackwall_base_url}"
}
}
},
"inputs": [
{
"id": "blackwall_api_key",
"type": "promptString",
"description": "Blackwall api key",
"password": true
},
{
"id": "blackwall_base_url",
"type": "promptString",
"description": "Blackwall base url",
"password": true
}
]
}{
"mcpServers": {
"blackwall": {
"command": "npx",
"args": [
"-y",
"blackwall-mcp@latest",
"blackwall-mcp"
],
"env": {
"BLACKWALL_API_KEY": "<blackwall_api_key>",
"BLACKWALL_BASE_URL": "https://..."
}
}
}
}file .mcp.json (in your project root)file claude_desktop_config.jsonfile ~/.cursor/mcp.jsonfile .vscode/mcp.jsonfile ~/.codeium/windsurf/mcp_config.jsoninstall npx -y blackwall-mcpversion 1.4.2
What it does
Blackwall is a risk-gating system for AI agents that evaluates irreversible actions (payments, SQL, deletions, emails) before they execute. It returns a risk score, reversibility assessment, and a GO/CAUTION/STOP recommendation. It's for developers who want to add a safety layer to agent-based systems.
- Stop an agent from running a DELETE query with no WHERE clause before the database is damaged.
- Require human confirmation when an agent attempts to send an email or transfer money above a threshold.
- Log all risky agent actions in observe mode to audit what the agent would have done, then enforce gates once confident.
Tools this server exposes
| Tool | Does |
|---|---|
| forecast | Pre-action risk assessment that evaluates an intended action (send_email, make_payment, run_sql, delete_file, post_content) and returns a recommendation (GO/CAU |
| observe | Post-action outcome report that closes the loop by telling Blackwall what actually happened after an action ran or was blocked, used to track prediction accurac |
Before you enable it
An MCP server runs with whatever access you give it. Start with a development account rather than production, and check which tools the server exposes before you let an agent call them unattended.