Brutor CLI

brutor is a small command-line tool for developers working against a Brutor gateway. Log in once — the session token lands in your OS keychain, never a dotfile — then discover the models and MCP servers your resource groups can use and wire your local AI tools (Claude Code, Cursor, an SDK script) to the governed gateway without hand-copying URLs.

It talks only to the Core Proxy’s end-user Portal API (/v1/portal/*), the same surface the User Portal uses. It changes nothing on the gateway — it’s a client, not an admin tool.

Install

cargo install --path brutor-cli      # builds the `brutor` binary
brutor --help

Log in

brutor login prompts for the gateway URL, tenant, username, and password. Accounts with TOTP two-factor are prompted for the code. The resulting portal JWT is stored in the OS keychain (Keychain on macOS, Credential Manager on Windows, Secret Service on Linux); only the non-secret connection settings go to ~/.config/brutor/config.json.

brutor login
#  Gateway URL [https://gateway.acme.com]:
#  Tenant ID [default]:
#  Username [ada]:
#  Password: ********
#  Logged in to https://gateway.acme.com (tenant default) as ada — session stored in the OS keychain.

brutor whoami

For CI or scripts, pass the password on stdin so nothing touches the TTY or a file:

printf '%s' "$BRUTOR_PASSWORD" | brutor login \
  --gateway gateway.acme.com --tenant default --username ci --password-stdin

Headless machines, containers and CI

The keychain needs a desktop session — on Linux specifically, a running Secret Service (GNOME Keyring / KWallet). A headless server or a bare container has none, so the CLI falls back to a 0600 file next to the config and prints a note saying it did. Nothing silently downgrades.

Two environment variables cover the fully non-interactive case:

Variable	Effect
`BRUTOR_TOKEN`	A portal JWT to use directly. Takes precedence over any stored session — no keychain, no file.
`BRUTOR_CONFIG_DIR`	Relocates all CLI state (default `~/.config/brutor`). Useful for containers and per-job CI isolation.

export BRUTOR_CONFIG_DIR=/var/lib/brutor
export BRUTOR_TOKEN="$(vault read -field=token secret/brutor/ci)"
brutor models --json

Resolution order for a session is BRUTOR_TOKEN → OS keychain → fallback file.

Discover models

brutor models lists the models your groups can use — the same access policy the portal and API keys enforce. --snippets prints ready-to-paste client configuration.

brutor models
# MODEL                                    PROVIDER         MODES
# OpenAI GPT-5.5                           openai           responses
# OpenAI GPT-5.2 Auto-Routing              openai           chat
# ...

brutor models --snippets

The snippets cover both endpoint families the gateway serves:

# OpenAI-compatible clients (SDKs, LangChain, LiteLLM, Open WebUI …)
export OPENAI_BASE_URL=https://gateway.acme.com/v1/proxy/llm
export OPENAI_API_KEY=<your Brutor API key>

# Anthropic-native clients (Claude Code, Anthropic SDK)
export ANTHROPIC_BASE_URL=https://gateway.acme.com
export ANTHROPIC_AUTH_TOKEN=<your Brutor API key>

See Coding agents through the gateway for the full client story.

Discover and connect MCP servers

brutor mcp search lists the governed MCP servers available to you; brutor mcp connect <name> prints the exact command to wire one into a client.

brutor mcp search github
# SERVER                       DESCRIPTION
# GitHub                       repos, issues, pull requests

brutor mcp connect github
# GitHub — governed MCP endpoint
# https://gateway.acme.com/v1/proxy/mcp/mcp-system-github-default
#
# # Claude Code
# claude mcp add --transport http github \
#   https://gateway.acme.com/v1/proxy/mcp/mcp-system-github-default \
#   --header "Authorization: Bearer <your Brutor API key>"
#
# # Generic MCP client config (JSON)
# {"mcpServers": {"github": {"type": "http", "url": "…", "headers": {…}}}}

Every tool call through that endpoint is governed — guardrails, per-tool limits, approvals, and audit — exactly like any other MCP traffic. See Use MCP tools through the gateway.

Install a skill into a local agent

Agent Skills are normally consumed server-side via progressive disclosure over MCP. For skills an admin has marked exportable, brutor skills downloads the bundle and unpacks it into a local coding agent’s skills directory.

brutor skills search              # skills your groups may export
brutor skills install pr-review   # → ~/.claude/skills/pr-review/
brutor skills update              # re-install tracked skills that changed

install writes SKILL.md, scripts/, references/, and assets/ into ~/.claude/skills/<name>/ (override with --dir). Only skills with local export enabled and accessible to your resource groups are listed or downloadable; every export is audited server-side. See Agent Skills → Local export.

One-step setup

brutor setup detects the AI tools installed on your machine and prints the wiring for each, so a fresh laptop is pointed at the governed gateway in one command.

brutor setup
# Detected local AI tooling:
#   ✓ Claude Code
#   ✓ Cursor
#   ✓ VS Code
# Point them at the governed gateway:
#   export OPENAI_BASE_URL=…
#   export ANTHROPIC_BASE_URL=…
# MCP: run `brutor mcp search`, then `brutor mcp connect <name>`.

Command reference

Command	What it does
`brutor login`	Authenticate; store the session in the OS keychain. `--password-stdin` for CI.
`brutor logout`	Remove the stored session.
`brutor whoami`	Show the logged-in identity and gateway.
`brutor models [--snippets] [--json]`	List usable models; optionally print client config.
`brutor mcp search [query] [--json]`	List/search governed MCP servers.
`brutor mcp connect <name>`	Print the connect command/config for one server.
`brutor skills search [query]`	List skills your groups may export.
`brutor skills install <name> [--dir]`	Download + unpack a skill into a local agent.
`brutor skills update`	Re-install tracked skills with a newer version.
`brutor setup`	Detect local AI tools and print gateway wiring.