Claude Desktop · json
{
"mcpServers": {
"intertrace-secured": {
"url": "https://intertrace.fly.dev/mcp",
"headers": {
"Authorization": "Bearer itr_rt_YOUR_RUNTIME_KEY",
"X-Intertrace-MCP-Server": "YOUR_SERVER_KEY"
}
}
}
}
# Optional inference gateway URL: https://intertrace.fly.dev/anthropicMCP remote URL; optional third-party inference via /anthropic.
+Three steps
- 01
Get a runtime key
Create an itr_rt_* key in Connect. It carries your tenant, so no org headers are needed.
- 02
Point the client at Intertrace
Paste the snippet. Nothing else in your code changes.
- 03
Watch decisions arrive
Every call shows up in Traffic with its verdict and the reason, in monitor mode first.
+What you get
Once it's connected
- Every tools/call authorized against policy before it reaches the server
- Tool descriptions inspected for hidden instructions
- Tool schemas pinned — drift is caught, not trusted
- Denied calls fail closed with a logged reason
+Related
+ The AI watching your AI
Protect your Claude Desktop agents today.
Start in monitor mode, see every decision, then turn on enforcement when you're ready.