okf-mcp since the early days of OKF

[mfdaves]

I started working on okf-mcp shortly after the Open Knowledge Format appeared.

That is what okf-mcp provides today.

It can:

• expose OKF knowledge directly to agents through MCP;
• validate bundles against OKF, including the new v0.2 trust signals;
• search concepts and navigate their relationships;
• build and inspect the knowledge graph;
• show provenance, lifecycle status and source information;
• work with local folders, remote Git bundles and larger federated projects;
• support proposals and controlled authoring without silently changing files;
• keep Markdown and Git as the source of truth.

It is local-first and does not require a vector database, an embedding service or a cloud account. Reading is safe by default, while every write requires explicit authorization.

The upcoming 0.9.0 release also introduces a generic producer system. A producer can extract knowledge from an external system - such as a PostgreSQL database - and return OKF documents. Before anything is written, okf-mcp previews and validates the result, checks ownership, protects existing files and ensures that the generated knowledge conforms to OKF v0.2.

I’ve tried to keep the project focused: it is not another knowledge platform. It is a practical bridge between OKF documents and the agents that need to discover, understand and maintain them.

I’d really value feedback from others using or experimenting with OKF - particularly around interoperability, producer integrations and the tools agents need when working with real knowledge bases.

Repository: https://github.com/mfdaves/okf-mcp