Kanonik
Governance runtime for compliance work, connected to Grok over MCP.
Kanonik gives your AI a typed compliance model to work in, a tamper-evident
record of everything it does, a server-side Verifier that checks its work, and a
human-approval gate in front of every write. Your AI does the reasoning. Kanonik
makes the result defensible.
What you can do with it
Ask your AI to work on your compliance program in plain language:
- Draft and revise policies and procedures against a loaded framework
- Assess internal controls and record the evidence behind each judgement
- Find the gaps between what a framework requires and what you actually have
- Build an audit trail an assessor can verify independently
- Produce a Statement of Applicability, or the equivalent artifact for your framework
The framework in scope comes from your workspace. Kanonik runs ISO 27001:2022,
SOC 2, GDPR, NIST CSF 2.0 and HIPAA today. Nothing in the plugin is specific to
any one of them.
Installing
Then pick Kanonik. Or from the CLI:
grok plugin install kanonik --trust
On first use your AI connects to https://app.kanonik.ai/mcp and a browser
window opens for sign-in. You will need a Kanonik account; sign up at
kanonik.ai.
What this plugin contains
One file that tells Grok where the Kanonik MCP server lives:
{
"mcpServers": {
"kanonik": { "type": "http", "url": "https://app.kanonik.ai/mcp" }
}
}
That is the whole plugin. There is no bundled code, no install script, no hook,
and no local process. Everything runs server-side at Kanonik.
Permissions and data
- Network. One endpoint,
https://app.kanonik.ai/mcp. Nothing else.
- Authentication. OAuth 2.1 with Dynamic Client Registration and PKCE,
against
https://auth.kanonik.ai. Tokens are held by your Grok client. The
plugin holds no credential and ships no secret.
- Scopes.
kanonik:read to read your compliance record, kanonik:propose
to draft changes, kanonik:commit to submit them for approval.
- Writes. No change reaches your record without a signed single-use approval
token and a human clicking to approve it. Your AI cannot approve its own work.
- Your other systems. Kanonik never holds credentials to your source systems.
If your AI reads from GitHub, Jira, or your identity provider, it does so
through its own connections, not through Kanonik.
Links
License
See LICENSE. The Kanonik service itself is proprietary; this
repository contains only the connection manifest and documentation for it.