agntcy/oasf

Open Agentic Schema Framework (OASF)

What it is – OASF is a community‑driven, open‑source schema system for describing AI agents. It defines a JSON‑based data model (the record object) that can capture an agent’s capabilities, skills, domains and extensible modules. The schema is versioned and immutable, so tools that rely on a particular version stay stable.

Why it matters – Modern AI‑agent ecosystems (autonomous assistants, tool‑using bots, RAG pipelines, etc.) need a common language to advertise what they can do and to discover each other. OASF provides that lingua‑franca, enabling interoperability across services, IDE plugins, and catalogues.

Key components

• Schema definitions – JSON files under schema/ (e.g., record.json) plus a taxonomy of skills and domains.
• Schema Server – An HTTP server (server/) that hosts the latest schema, offers a UI for browsing, and validates records via a REST API.
• Tooling – A Taskfile‑driven workflow for building, testing, and deploying the server with Docker/Kind. Hot‑reload support lets you iterate on schema changes without restarting the whole stack.
• MCP integration – The Directory MCP Server can be hooked to an LLM (e.g., Cursor) to generate, validate, and convert OASF records directly from code.

Typical workflow

1. Write a record – Create a JSON record describing an agent, annotating it with skills and domains.
2. Validate – Send the JSON to the OASF server (/validate endpoint) or use the MCP server’s LLM‑assisted commands.
3. Publish / discover – Other services can query the server for records that match a skill or domain, enabling dynamic discovery.

Getting started

# Clone
git clone https://github.com/agntcy/oasf.git && cd oasf

# Install prerequisites (Taskfile, Docker, Go, yq, curl, tar)
# Build dependencies and artifacts
task deps
task build

# Deploy a local server (creates a Kind cluster, Helm‑installs the service)
IMAGE_TAG=latest task build:server
task up   # then open http://localhost:8080

# For hot‑reloading schema changes
task reload

Extending the schema – You can add private extensions (new object types, extra fields) and host a custom server instance. The contribution guide explains how to propose new skills, domains, or modules and how to version them.

Where to find more

• Live schema UI: https://schema.oasf.outshift.com
• Taxonomy of AI Agent Skills: https://schema.oasf.outshift.com/skill_categories
• Protocol‑buffer definitions (for generating language SDKs): https://buf.build/agntcy/oasf
• MCP server integration: https://github.com/agntcy/dir/tree/main/mcp

Who might use it

• Teams building fleets of autonomous agents that need a shared capability catalog.
• Platform providers offering a marketplace of AI agents.
• IDE plugin developers who want LLM‑assisted generation of agent metadata.
• Researchers standardising agent benchmarks and reproducibility.

---

All information above is taken directly from the repository’s README; no external assumptions have been added.