Integrations Overview

Behavry sits inline between your AI surfaces and the tools / models they call. Existing clients connect by updating an MCP server URL, an API base URL, or installing the browser extension — no SDK changes required.

Supported clients

MCP clients (IDEs, editors, CLIs)

These connect by adding Behavry as an MCP server (or HTTP MCP URL) in their config.

Client	Type	Setup
Claude Code	CLI	Guide
Claude Desktop	Desktop app	Guide
Cursor	IDE	Guide
Zed	Editor	Guide
Windsurf	IDE	Guide
Google Antigravity	IDE	Guide
OpenAI Codex CLI	CLI	Guide
Amazon Kiro	IDE	Guide
Open Interpreter	CLI	Guide

LLM API proxies

These point your API base URL at Behavry, which then forwards to the upstream provider while inspecting the request and the response.

Provider	Setup
Anthropic API (Claude)	Guide
OpenAI API (GPT)	Guide
Google Gemini API	Guide
Ollama API (local)	Guide
NVIDIA NemoClaw	Guide
NVIDIA OpenShell	Guide

Browser surfaces

One Chrome extension covers every major AI web app.

Surface	Covered via
ChatGPT, Claude, Gemini, Copilot, Perplexity, Poe, Mistral, Meta AI, HuggingChat, DeepSeek, Grok, Phind	Browser Extension

At the time of writing the extension covers 12 AI services, but the fingerprint database is updated regularly — see the extension guide for the current list.

How the inline proxy works

Your agent / IDE / API client
        │
        ▼
    Behavry
  ┌──────────────────────────────┐
  │ identity  → policy  → DLP    │
  │ context gate → inbound rules │
  │ audit + siem + metrics       │
  └──────────────────────────────┘
        │
        ▼
  Target MCP server / LLM API

Every in-flight call is authenticated, evaluated against OPA policy, scanned for sensitive content, rolled into behavioral baselines, and logged. For MCP clients, tools/list responses also pass through Context Window Governance.

Automated setup (local)

make setup      # provisions an agent and writes all supported client configs
make tokens     # refresh expired tokens

make setup detects which of the supported clients are installed on your machine and rewrites their MCP config to point at Behavry. Run make setup --client=claude-code to target only one.

Manual setup (any client)

1. Register an agent in the dashboard (Agents → New agent) and copy the token
2. Set the MCP server URL — for MCP clients, point at http://localhost:8000/mcp/v1; for LLM proxies, point the provider base URL at http://localhost:8000/{provider}/v1
3. Add an Authorization: Bearer <token> header to the client config

Tool calls from that client are now governed.

Provisioning at scale

The Agents → Provisioning page supports:

• Single provision — one click, pick client type, get a token + pre-filled config
• Bulk provision — enroll multiple agents from a form
• CSV upload — pre-fill from a CSV (name, type, role, tags) for large rollouts
• Test connection — live SSE stream of the first handshake to confirm the client is reachable

See the Provisioning tab under Agents in the dashboard, or POST /api/v1/agents/provision.

Related

• Agent Identity — the identity layer
• Context Window Governance — tool filtering for MCP clients
• Browser Extension — coverage for the 12 browser AI surfaces
• Human AI Governance — extends governance to humans using AI in a browser