Skip to main content

Other Enrollment Methods

Beyond agents, IDEs, and browsers, four more methods extend coverage to servers, SaaS AI, raw model traffic, and whole endpoints. Each is a card in the Provisioning workspace.

MCP Server Registration​

Register an MCP server — HTTP or stdio — and Behavry inserts itself between the agent and that target, mediating every tool call. This gives you a tool-surface audit, DLP redaction, and context gating for any MCP server your agents use.

behavry mcp register --url https://your-mcp-server.example.com ...

SaaS Connector Activation​

OAuth Behavry into a SaaS AI product — Slack AI, GitHub Copilot, Salesforce Einstein — to observe, redact, and gate that vendor's AI features. The flow is OAuth → policy bind → telemetry, giving per-connector coverage with vendor AI-feature audit and DLP redaction. See also AI Exposure.

SaaS platforms you can't sit in front of

For agent platforms where the vendor's cloud makes the outbound tool calls — Microsoft Copilot Studio being the common one — Behavry can't be a network chokepoint. Instead you make it the only sanctioned MCP connector, tenant-wide, via a Power Platform data policy. See Copilot Studio.

Docker AI, dev containers & WSL

Docker's Ask Gordon agent, dev containers, and WSL all govern the same way — route their MCP traffic through Behavry instead of Docker's MCP Gateway. Policy and Decision Traces follow the traffic, not the host; only endpoint exposure signal is namespace-bound. See Docker AI (Gordon).

Enforcement Proxy​

Point a model client's base URL at Behavry's inline enforcement layer to intercept every model call for pre-execution authorization and Decision Trace generation — with no SDK changes. You get pre-execution prompt audit, output DLP, and rate + spend guardrails.

# set the client base_url to:
https://proxy.behavry.ai/v1

Behavry Wardens​

Deploy the Behavry Wardens desktop agent (Windows / macOS) for OS-level TLS interception that governs every AI app and browser on the endpoint — full-endpoint coverage of all HTTP AI traffic, not just the browser, with device-bound identity and clipboard/screenshot context. Deploy by generating an enroll token and running the Warden agent on the device.

For install and upgrade steps, configuration, ports, and the full command reference, see Warden desktop agent.

These flows produce the same result

However an actor is enrolled, it ends up with an identity, a role set, and a stream of Decision Traces. From there, every operational surface treats every governed actor the same way.