Ai Agent Firewall MCP server
{
"mcpServers": {
"fidacy": {
"command": "npx",
"args": [
"-y",
"@fidacy/mcp@latest"
],
"env": {
"FIDACY_ENGINE_URL": "https://...",
"FIDACY_ENGINE_API_KEY": "<fidacy_engine_api_key>"
}
}
}
}{
"mcpServers": {
"fidacy": {
"command": "npx",
"args": [
"-y",
"@fidacy/mcp@latest"
],
"env": {
"FIDACY_ENGINE_URL": "https://...",
"FIDACY_ENGINE_API_KEY": "<fidacy_engine_api_key>"
}
}
}
}{
"mcpServers": {
"fidacy": {
"command": "npx",
"args": [
"-y",
"@fidacy/mcp@latest"
],
"env": {
"FIDACY_ENGINE_URL": "https://...",
"FIDACY_ENGINE_API_KEY": "<fidacy_engine_api_key>"
}
}
}
}{
"servers": {
"fidacy": {
"command": "npx",
"args": [
"-y",
"@fidacy/mcp@latest"
],
"env": {
"FIDACY_ENGINE_URL": "${input:fidacy_engine_url}",
"FIDACY_ENGINE_API_KEY": "${input:fidacy_engine_api_key}"
}
}
},
"inputs": [
{
"id": "fidacy_engine_url",
"type": "promptString",
"description": "Fidacy engine url",
"password": true
},
{
"id": "fidacy_engine_api_key",
"type": "promptString",
"description": "Fidacy engine api key",
"password": true
}
]
}{
"mcpServers": {
"fidacy": {
"command": "npx",
"args": [
"-y",
"@fidacy/mcp@latest"
],
"env": {
"FIDACY_ENGINE_URL": "https://...",
"FIDACY_ENGINE_API_KEY": "<fidacy_engine_api_key>"
}
}
}
}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 @fidacy/mcpversion 0.7.1
What it does
An MCP server that gates AI agent payment actions against cryptographically signed mandates before funds move. It provides both advisory trust verdicts and enforcement-layer payment authorization, with verifiable audit trails and non-custodial operation.
- Block payment actions (wrong payee, inflated amount, duplicate invoices) before execution by requiring a signed grant from the firewall.
- Get a signed trust verdict for a proposed action (payment, message, voice call) that anyone can verify against public keys without trusting Fidacy.
- Retrieve hash-chained audit proofs for payment decisions to demonstrate compliance and catch prompt-injection or hallucination attacks.
Tools this server exposes
| Tool | Does |
|---|---|
| assess_action | Returns a signed Fidacy trust verdict for a proposed action (payment, message send, voice call, or custom) with verifiable cryptographic proof. |
| request_payment | Authorizes a payment action against a signed mandate; returns either a short-lived Ed25519 grant (ALLOW) or a denial reason (DENY). |
| verify_mandate | Reads the mandate envelope and Fidacy public key to validate payment constraints. |
| get_audit_proof | Returns a hash-chained, verifiable proof for a payment decision by its ID. |
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.