Skills

Awesome-OKF: Connect the OKF resource catalog to your AI agent

Best fit if you want an AI agent to search the Open Knowledge Format (OKF) catalog of tools, plugins, skills, and documents, or to convert existing content in bulk into an agent-friendly OKF format. You need a Python environment and an MCP-capable client, and you must remember that README files are generated: change entries only in data/catalog.yaml.

Official source · Awesome-OKF

Install

pipx install awesome-okf

Who it is for

  • Developers or agent users who want an AI agent to query the OKF (Open Knowledge Format) resource catalog directly: the catalog includes OKF-related tools, plugins, skills, proposals, and documents.
  • People already using an MCP-capable client who want programmatic search with tools such as search_catalog and get_catalog_entry.
  • People who want to turn their own content into an OKF catalog (Markdown files + YAML frontmatter) and check it with myokf validate.

When it is not a fit:

  • Situations where you do not need OKF resource discovery or conversion and only want to query data in other domains.
  • Direct editors of the README — official docs state that README*.md files are generated, must not be hand-edited, and you must change data/catalog.yaml then regenerate.
  • Points the official docs do not cover (such as the supported client list, the input-language range for conversion, or concrete system requirements) are not guessed here; they are marked unconfirmed.

First try in 3 minutes

Prerequisite: a local Python environment with pipx installed (the official install method). This is a Python-implemented CLI + MCP meta-server.

  1. Install: run this in a terminal

    pipx install awesome-okf

  2. Connect the agent: register the MCP server with your MCP client using the command awesome-okf-server (the official note is "Connect it to your MCP client"; exact registration paths per client are not in the evidence, so follow your client documentation).

  3. Start a low-risk read-only task: tell the agent "use search_catalog to search the catalog for OKF tools and return the first few entries". The tool accepts query/kind/limit parameters, performs a catalog query, and does not write.

Success signal: the agent returns catalog entries (based on get_catalog_entry / list_catalog data) or catalog_stats totals — showing the meta-server is connected.

Note: the first try should only query. convert_to_okf and scripts/convert-to-okf.py create OKF catalog files and write to disk. Run them only after you confirm the target path and know they will produce Markdown + YAML frontmatter files.

Use cases

Case 1: Ask the agent to find OKF resources. Scenario: you want to see which tools or skills exist in the OKF ecosystem. Ask the connected agent: "use search_catalog, put your keywords in query, optionally filter by kind, and use limit for how many entries". The agent calls the meta-server search tool and returns matching tools/plugins/skills/proposals/documents; use catalog_stats for totals. This is query-only; the result is the catalog entries themselves.

Case 2: Turn existing content into an OKF catalog. Scenario: you have structured text (such as lists, JSON, key-value pairs, or URL lists) and want an agent-friendly knowledge format. Use scripts/convert-to-okf.py; its --format anything mode tries the parse chain (list → JSON → key-value → URLs → line-per-row) until one parser succeeds. What you can confirm: it produces a catalog of Markdown files with YAML frontmatter, which you can then check with myokf validate. Side effect: this step writes files to disk; conversion runs on a local engine. Confirm the output directory first. Not covered by evidence: whether calling the convert_to_okf MCP tool does the same conversion, and the exact error behavior when conversion fails.

Limits

  • Do not edit README*.md directly: they are generated. To change entries, edit data/catalog.yaml and regenerate; official guidance requires entries to pass awesome-okf validate and pytest before you send them.
  • Network/file access: conversion uses a local engine (extraction done by local engines only) and does not fetch from outside. Official docs do not describe the meta-server's own network behavior; mark that as unknown.
  • Confirm first: any action that writes files (conversion, generation) should confirm the target path first. Query tools (search_catalog, list_catalog, get_catalog_entry, catalog_stats) read the catalog, but evidence is not strong enough for an absolute safety claim of "does not write files or run commands", so this page does not assert that.
  • Runtime: a Python-implemented CLI and MCP meta-server installed via pipx; you need an MCP-capable client. Official docs do not name which MCP clients are supported.
  • Unconfirmed: the supported natural-language range, per-client registration details, and system dependency versions are not in the official text. Read the latest repository docs before use.