Why OKP is skills-first — and why we chose skills over MCP #1
markbang
announced in
Announcements
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Why OKP is skills-first — and why we chose skills over MCP
When we built OKP (Open Knowledge Protocol), we had to decide how AI agents would consume knowledge. Two integration paths were on the table: MCP servers and agent skills. We went skills-first, and here is the reasoning behind that call.
The problem with MCP for knowledge
MCP (Model Context Protocol) is a reasonable tool-transport standard, but as a distribution layer for knowledge it carries real costs:
Why skills fit knowledge better
A skill is a Markdown instruction file — the agent reads it and learns how to work with a domain:
okp-searchstarts by reading the domain README, then filters by schema, then follows links between concepts. The workflow is the knowledge.npx skills add, run anywhere your agent runs. No server, no token, no process to babysit.SKILL.mdis documentation, executable instruction, and searchable content in one file. The same file that teaches your agent is the page search engines index.What OKP chose
OKP ships two skills —
okp-searchandokp-import— and nothing else. No MCP server. The protocol itself is open and self-hostable; skills are the only integration surface.The philosophy in one sentence: the context layer for AI agents should be knowledge you can read, not a service you must run.
Try it
Questions, pushback, or your own MCP war stories — drop them below.
All reactions