aevrin

Aevrin MCP Security Scanner CLI. Wraps the same open-source scanner binaries and normalization logic (aevrin-scanner-core) that the Aevrin backend uses, run locally against your own machine. Results save to your Aevrin dashboard automatically once you're logged in, pass --no-upload for a purely local, ephemeral scan.

Install

python3 -m pip install --upgrade aevrin

The same CLI is also available through npm:

npm install --global aevrin

Requires Docker using Linux containers: a scan starts the MCP server inside one disposable, version-pinned engine image, which is pulled automatically when missing. On Docker Desktop, assign at least 4 GB of memory and permit bind mounts from the system temporary directory. There is no non-Docker fallback - an unavailable sandbox is an unavailable scan, never a scan performed without one - so use --remote for a local folder if you would rather not run Docker.

Usage

aevrin scan mcp "npx -y @playwright/mcp"   # scan a server by its launch command
aevrin scan ./my-mcp-server
aevrin scan github.com/owner/repo
aevrin scan https://my-live-server.example.com --json
aevrin scan ./my-mcp-server --fail-on high
aevrin scan ./my-mcp-server --no-upload   # skip saving to your dashboard (e.g. in CI)

Target type is auto-detected: a github.com URL, another public https:// URL as a live MCP server, or anything that exists on disk as a local path. The same checks run either way - one engine over the tools a live server actually returns, in five stages (resolving, launching, enumerating, analyzing, grading) - and the target type only changes how resolving works out what to start. Private, loopback, metadata, credential-bearing, and plain-HTTP live targets are rejected. Aevrin never executes submitted stdio MCP commands.

Flags

Exit codes

Results go to stdout; stage progress and diagnostics go to stderr, safe to pipe --json output without stage-progress noise mixed in.

Aevrin prints one grade letter and a risk score, then each finding as its own block rather than a table row, because a table forced every finding down to a title. Stage progress is listed above it. Run aevrin scan --help, or see the full reference for the exact shape: https://docs.mcp.aevrin.net/cli.

Other commands

aevrin login / logout           # browser device-code login, credentials in ~/.aevrin
aevrin agent scan               # what the AI coding agents on this machine may do
aevrin agent scan --json        # versioned snapshot; never contains a credential value
aevrin findings triage <id> <status> [--reason ...]
aevrin mcp-server               # local stdio MCP server with one tool, scan_mcp_server
aevrin mcp-header               # print the Authorization header for Aevrin MCP
aevrin registry publish <dir>   # Aevrin administrators only

mcp-server needs the MCP extra: pip install "aevrin[mcp]", and registry publish the YAML extra: pip install "aevrin[registry]".

Aevrin MCP (https://api.mcp.aevrin.net/mcp) needs your Aevrin API key. mcp-header prints {"Authorization":"Bearer <key>"} from the key aevrin login stored, so Claude Code can read it without the key being written into its config:

claude mcp add-json --scope user aevrin '{"type":"http","url":"https://api.mcp.aevrin.net/mcp","headersHelper":"aevrin mcp-header"}'

agent scan reads configuration only: no agent is started and nothing from a config file is executed. Nothing leaves the machine without --upload. Full reference: https://mcp.aevrin.net/docs/cli.

Development

uv sync
uv run pytest tests -v
uv run ruff check .
uv run mypy aevrin_cli