Hub
AI agents and agent tooling
What separates an agent from a model, and the open protocols, SDKs, and coding agents used to build agents, including the Model Context Protocol, A2A, and AGENTS.md.
A model takes an input and returns an output. An agent is software that wraps one or more models with instructions and tools so that it can take actions. The OpenAI Agents SDK, for example, defines agents as models configured with instructions, tools, guardrails, and handoffs to other agents, and its tools include functions, MCP servers, and hosted tools. Google's Agent Development Kit describes itself as a framework for building, evaluating, and deploying agents that is optimized for Gemini but model-agnostic.12
The agent and the model are often separate pieces with separate terms. Codex CLI, OpenAI's Apache-2.0 coding agent, runs on your computer but signs in with a ChatGPT plan or an API key to use OpenAI's models. goose, an agent with a desktop app, a command line, and an API, works with several model providers, among them Anthropic, OpenAI, Google, and Ollama. An open-source agent can therefore depend on a hosted model service.34
Three open specifications connect agents to tools, to each other, and to codebases. The Model Context Protocol (MCP) uses JSON-RPC 2.0 messages between hosts (LLM applications), clients inside them, and servers that offer resources, prompts, and tools. Agent2Agent (A2A) lets agents built on different frameworks by different companies discover each other's capabilities through Agent Cards and work together without exposing their internal memory or tools. AGENTS.md is a Markdown file in a project that gives coding agents instructions; its maintainers call it a README for agents. All three are projects of the Agentic AI Foundation, a Linux Foundation project.5678
Because agents act, permissions matter. The MCP specification treats tools as arbitrary code execution, says hosts must obtain explicit user consent before invoking any tool, and notes that the protocol itself cannot enforce these principles. It asks implementers to build consent and authorization flows into their applications. When you assess an agent, look at how it asks for approval, what it is allowed to run, and what data it sends where.5
What this hub covers
Covers software that uses models to take actions: agent protocols and file conventions, agent SDKs and frameworks, coding agents, and benchmarks that test agents. The models that agents call are separate records with their own licenses. It does not cover hosted agent products in depth or compare agent performance. Featured records are examples chosen to cover the subject, not a ranking or a complete list.
Reading path
- Agents and roboticsStart here for how agents and robots differ from the models they use.
- How the ecosystem fits togetherWhere agents sit relative to models, runtimes, and compute providers.
- Hosted APIMany agents call a model through a hosted API. This explains what that means.
- Model Context Protocol recordThe catalog record for MCP, including its governance and license transition.
- Agentic AI FoundationThe Linux Foundation directed fund that hosts MCP, A2A, AGENTS.md, and goose.
- Hosted or local?Where the model behind an agent runs affects what data leaves your machine.
- Reading evaluationsHow to read agent benchmark results, such as Terminal-Bench, with care.
In the catalog
- Organizations working on agentsOrganizations whose records list agents as a sector.
- Agent protocols, SDKs, and toolsOpen artifacts that match agents, including protocols, SDKs, and coding agents.
- Coding agents and related toolsA text search that matches records tagged as coding agents.
Featured records
Examples chosen to cover the subject; not a ranking or a complete list.
- Agentic AI Foundation (AAIF)Organization
- Model Context Protocol (MCP)Open model or tool
- Agent2Agent (A2A) ProtocolOpen model or tool
- AGENTS.mdOpen model or tool
- OpenAI Agents SDKOpen model or tool
- Agent Development Kit (ADK)Open model or tool
- smolagentsOpen model or tool
- Codex CLIOpen model or tool
- gooseOpen model or tool
- Cline (open-source coding agent)Open model or tool
- Terminal-BenchOpen model or tool
Primary documents
- Model Context Protocol specification (revision 2026-07-28) (external site: modelcontextprotocol.io)Model Context ProtocolThe normative MCP specification, including the consent, privacy, and tool-safety principles that implementers are expected to follow.
- Agent2Agent (A2A) Protocol README (external site: raw.githubusercontent.com)A2A Project (GitHub)The A2A project's own summary of agent discovery through Agent Cards and its Linux Foundation stewardship and Apache 2.0 license.
- AGENTS.md README (external site: raw.githubusercontent.com)AGENTS.md (GitHub)Shows what an AGENTS.md file contains and why it is a convention rather than software.
- AAIF projects (external site: aaif.io)Agentic AI FoundationThe Agentic AI Foundation's list of hosted projects, useful for checking which agent projects are under neutral stewardship.
- OpenAI Agents SDK README (external site: raw.githubusercontent.com)OpenAI (GitHub)A concise statement of the parts of an agent (instructions, tools, guardrails, handoffs, tracing) as one SDK defines them.
- Codex CLI README (external site: raw.githubusercontent.com)OpenAI (GitHub)Shows that a locally running, open-source agent can still require a hosted model account or API key.
- goose README (external site: raw.githubusercontent.com)Agentic AI Foundation, aaif-goose (GitHub)Describes an agent that can use different model providers and that is now part of the Agentic AI Foundation.
Sources · reviewed Oct 1, 2026
Support Us
Help keep USASI useful.
Optional. No USASI account required. Payment takes place on the linked provider’s website (Buy Me a Coffee).
About supporting this project