Skip to main content

Behavry Integration — Open Interpreter

Open Interpreter connects to Behavry's unified MCP gateway (/mcp/v1) over Streamable HTTP, authenticating with the agent's client credentials as an HTTP Basic header. Because Open Interpreter has unrestricted shell access, Behavry classifies it at critical risk tier by default.

Prerequisites​

  • A running Behavry instance (your control-plane URL, or http://localhost:8000 for a local stack)
  • Open Interpreter installed
  • An enrolled agent (gives you the client_id / client_secret)

1. Enroll the agent​

In the dashboard: Agents → Enroll Agent → Open Interpreter. Behavry provisions the agent and shows the config block with credentials filled in.

2. Config block​

Add to ~/.openinterpreter/config.yaml:

mcp_servers:
  behavry:
    url: https://<your-behavry-host>/mcp/v1
    transport: streamable-http
    headers:
      Authorization: "Basic <base64(client_id:client_secret)>"

Risk Tier: Critical​

Open Interpreter can execute arbitrary shell commands and write to any file. By default:

  • All shell commands are logged with full audit metadata
  • Destructive operations trigger HITL escalation
  • File writes to sensitive paths are blocked
  • DLP scanning runs on all output

Adjust the risk tier from the dashboard after enrollment if needed.

Local vs. Remote LLMs​

Behavry governs the tool-execution layer regardless of the LLM backend. To also audit LLM API calls, configure the corresponding Behavry API proxy (OpenAI, Anthropic, Ollama, or Gemini) alongside this integration.

Verify​

  1. Start Open Interpreter: interpreter
  2. Ask it to perform an action
  3. The tool call appears in the dashboard → Live Activity with a policy result

Troubleshooting​

401 Unauthorized — Credentials wrong or agent revoked. Re-enroll and paste the fresh block.

Server not connecting — Confirm the url ends in /mcp/v1 and the backend is reachable: curl https://<your-behavry-host>/health.