Skip to content
toolnexus

Ecosystem & references

toolnexus doesn’t invent new standards; it unifies existing ones behind a single Tool interface. Here’s what it builds on and where to go deeper.

The standards

Section titled “The standards”
  • Model Context Protocol (MCP) — the open standard for connecting models to tools, resources, and prompts over JSON-RPC (local stdio + remote HTTP). toolnexus reads an mcp.json and exposes every server. See MCP servers.
  • Agent Skills — the SKILL.md folder format with progressive disclosure. toolnexus globs a skills/ folder and exposes one skill tool. See Agent skills.
  • Agent2Agent (A2A) — the protocol for agents to call each other: an Agent Card advertises skills, JSON-RPC drives the exchange. toolnexus can both consume a remote A2A agent as a tool and serve your toolkit as one.
  • Model tool-call formats — the same uniform tools are emitted in OpenAI, Anthropic, and Gemini shapes, so you can drive any of them.

Inspiration

Section titled “Inspiration”
  • opencode — the project whose insight toolnexus ports: MCP tools, agent skills, your own functions, HTTP endpoints, and remote agents are the same thing to an LLM. toolnexus makes that idea vendor-neutral and byte-identical across five languages.

The contract

Section titled “The contract”

Everything above is pinned in SPEC.md in the repository — the one-page cross-language conformance contract that guarantees the five ports behave identically. The same examples/ fixtures (one mcp.json, one skill) produce the same output in JavaScript, Python, Go, Java and C#. That parity is the product.