Skip to content

MOCA, OKF and MCP

Three open efforts sit close together. They do different jobs, and work together.

What it is Its job
OKF, the Open Knowledge Format A file format: Markdown pages with YAML frontmatter and trust fields How one page of knowledge is written
MOCA, Modular Ontology, Content & Attestation A package format and the Readers that read it How a body of knowledge travels: with its version, structure, sources and signatures, read the same way everywhere
MCP, the Model Context Protocol A protocol between AI applications and the tools and data they use How an agent asks for knowledge at run time

MOCA builds on OKF; it does not replace it. Every page in a MOCA package is an OKF document, and MOCA reads OKF’s own fields (type, generated, verified, status, stale_after) as OKF defines them.

OKF stops at the file, on purpose. MOCA adds what a body of knowledge needs when it moves between systems or organisations: a package identity and version, one digest over every file, signatures and reviews, the original sources inside the package, validity windows, audiences, translations, composition from other packages, and an optional ontology. A MOCA package without any of these is still a valid OKF bundle with a manifest (why MOCA).

MCP is how an agent reaches knowledge; MOCA is the knowledge it reaches. The MOCA MCP server, moca-mcp, serves packages to any MCP client:

  • tools to list packages, search them, fetch a page and ask structure questions;
  • every page as an MCP resource, so generic clients can read packages too;
  • the same trust rules as every Reader: signatures verified against your trust root, and an audience filter the caller cannot widen.

So an MCP-capable assistant gets citations it can check, and content that is current and meant for it, without any MOCA-specific code. See the MCP server (one download, nothing to install) and the walkthrough.

Write pages in OKF, ship them as a MOCA package, and serve them over MCP, or read them in-process with a MOCA Reader.