Your documents, exposed as MCP tools — for your AI agents and expert workflows.
One MCP endpoint per K-AI instance. OAuth + scopes. Audit for experts (cleaning, conflicts), Retrieval for consumers (sourced answers). Compatible with Claude Desktop, Cursor, Copilot Studio.
The DKP, in practice.
K-AI implements the DKP across three surfaces — Audit, Platform, MCP — that operate the five layers.
K-AI MCP — one MCP server per role
K-AI exposes its functions via Model Context Protocol. Experts drive cleanup from their IDE or Claude Desktop; AI agents consume the cleaned base with no custom integration.
Wired into Claude Desktop or Cursor: the Steward asks in natural language for open conflicts in their domain, and handles remediations without leaving their tool.
// MCP tool call from the expert’s IDE/AI { "tool": "audit.list_conflicts", "arguments": { "domain": "hse-europe", "severity": ">= warning", "limit": 5 } }
{
"conflicts": [
{
"topic": "Intervention delay on sensor alert",
"documents": [
{ "id": "POL-HSE-2019", "extract": "… intervene within 4 hours …" },
{ "id": "DS-2024-014", "extract": "… intervention within 30 minutes …" }
],
"status": "open"
}
],
"total": 1
}Who is it for?
K-AI MCP Audit targets Document Stewards and Producers from their IDE. K-AI MCP Retrieval targets Document Consumers — humans through copilots, AI agents through tool calls.
See the relevant roles →Documentation
Full MCP spec, OAuth scopes, examples for Claude Desktop / Cursor / in-house agents, tool call schemas — on GitBook.
GitBook documentation →