What can your agents already reach?

Paste an MCP client config — Claude Desktop, Cursor, VS Code, a .mcp.json — and get a deterministic Reach × Capability × Consequence score for every path to money, data, production, or machines. No account, no model reads it, and the same input always gives the same number.

Stored so you can come back to the result. Nothing is sent to a model, and we never connect to your servers from here — only the CLI does that, from your machine.

The result appears here. A config with two write-capable servers and no gate usually scores in the 40s.

Or run it where the config lives

npx @interlok/cli scan                 # finds local MCP configs
npx @interlok/cli scan --json          # machine-readable
npx @interlok/cli scan --format cef    # into your SIEM
npx @interlok/cli scan --ci --max-critical 0   # fails the build

The CLI connects to each server, lists its tools, hashes the descriptions, and grades the write paths. It sends nothing anywhere unless you pass --upload.

What the score means

A · 90+Write paths gated, determinations human and IdP-backed, ledger anchored.
B · 75Gated, but identity or anchoring is missing.
C · 60Some write paths are still outside the policy boundary.
D · 40Multiple ungated write paths and no record of what ran.
FNothing between an LLM and a consequence.

Fixing the findings is the product. Start free and the first hold can be opened once your proxy and policy are connected.