What is the difference between OKF and MCP?

OKF (Open Knowledge Format) is the format your knowledge is packaged in — graph-linked Markdown with YAML frontmatter. MCP (Model Context Protocol) is the protocol an AI agent uses to connect to a source and fetch context at runtime. OKF is the data at rest; MCP is how the agent reaches it.

Do OKF and MCP compete?

No. They are complementary layers. OKF defines how knowledge is structured and made portable and verifiable; MCP defines how an agent connects to and retrieves that knowledge. Used together, an agent can traverse an OKF graph over MCP.

How does Claude use OKF over MCP?

CATAAM serialises your compliance program into an OKF bundle, and the Claude MCP connector exposes it to Claude. You ask a plain-English question, MCP carries the request, and OKF supplies the structured, current answer from your real graph.

← Blog

OKF

OKF vs MCP: How Open Knowledge Format and the Model Context Protocol Work Together

June 27, 2026 · 6 min read

OKF and MCP aren’t competitors — they’re a stack. The format your knowledge lives in, and the protocol an agent uses to reach it.

Open Knowledge Format (OKF) and the Model Context Protocol (MCP) come up in the same conversations, and it is easy to assume they compete. They do not — they sit at different layers and work best together.

OKF is the format. MCP is the protocol.

OKF is how knowledge is packaged — graph-linked Markdown with YAML frontmatter. MCP is how an AI agent connects to a source and pulls context at runtime. One is the shape of the data at rest; the other is the pipe the agent reads it through. Think "file format" vs "the connection that serves the file."

How they compose

Your knowledge is serialised into OKF — a verifiable, graph-linked bundle
An MCP connector exposes that bundle to an agent like Claude
The agent traverses the OKF graph over MCP and answers grounded questions
Because both layers are open, you avoid lock-in at the format and the transport

A concrete example

CATAAM serialises your compliance program into OKF, then the Claude MCP connector lets Claude read it. You ask "which controls cover this S3 bucket?" — MCP delivers the request, OKF supplies the structured, current answer from your OKF Context Engine. The format makes the data portable and verifiable; the protocol makes it reachable.

Which do you need?

Both. MCP without good context is a fast pipe to nothing useful; OKF without a protocol is a great bundle no agent is reading. Together they give an agent a verifiable knowledge graph it can actually query.

For more, read OKF for AI agents and how OKF is delivered through Git.

See OKF and MCP work together over your compliance graph

Explore the OKF Context Engine →

Frequently asked questions

What is the difference between OKF and MCP?: OKF (Open Knowledge Format) is the format your knowledge is packaged in — graph-linked Markdown with YAML frontmatter. MCP (Model Context Protocol) is the protocol an AI agent uses to connect to a source and fetch context at runtime. OKF is the data at rest; MCP is how the agent reaches it.
Do OKF and MCP compete?: No. They are complementary layers. OKF defines how knowledge is structured and made portable and verifiable; MCP defines how an agent connects to and retrieves that knowledge. Used together, an agent can traverse an OKF graph over MCP.
How does Claude use OKF over MCP?: CATAAM serialises your compliance program into an OKF bundle, and the Claude MCP connector exposes it to Claude. You ask a plain-English question, MCP carries the request, and OKF supplies the structured, current answer from your real graph.