For AI and agent builders

A compliance answer your agent can cite

An MCP endpoint over a judged control graph. The free tools need no key, so an agent is useful on its first call.

The job: Stop an agent producing plausible compliance text, and give it something it can point at.

Connect in one paste

Claude Code, Claude Desktop, Cursor and Windsurf configs, plus the packages on npm and PyPI. The crosswalk and catalogue tools answer unauthenticated.

Get the config

Ask the question a single-framework tool cannot

How much of a target standard a company already satisfies by holding another, in one direction, with the reasoning and the rejected claims returned alongside the number.

Run it in the browser

Check it in CI

A GitHub Action queries the same graph during a build and can fail a pull request on a gap, with the reasoning in the step summary.

See the developer surface

When this is the wrong tool

If you want a model that writes compliance prose, this is the wrong layer. It returns structured claims with provenance and expects your agent to do the writing.

Questions

Do I need an account to try it?

No. Anonymous callers get ten requests a day per IP with no key at all, which is enough to decide whether the answers are worth anything. A free account raises it, and the paid tier raises it again.

What stops the answers being made up?

The tools return records, not prose. A coverage number comes back with the controls it rests on, the reasoning for each, and the claims that were proposed for the same pair and rejected. If your agent cannot cite it, we did not return it.

The graph holds 686 frameworks and 59,123 judged cross-framework mappings, with 29,347 more that were judged and rejected and kept where you can read them.