Ask AsyncAPI

Answers about the AsyncAPI spec, concepts, and tooling — grounded in the official docs, with sources you can verify.

Use it in your own tools locally

The AsyncAPI knowledge base speaks MCP and ships as a PyPI package with the whole vector index bundled — it runs fully on your machine, no API keys, no remote calls. Wire it into your editor, install the Claude Code plugin, or clone the repo and run this whole chat locally.

Any MCP client — .mcp.json

{
  "mcpServers": {
    "asyncapi-knowledge": {
      "command": "uvx",
      "args": ["asyncapi-knowledge-mcp==0.0.2"]
    }
  }
}

Claude Code

# MCP server — one command
claude mcp add asyncapi-knowledge -- uvx asyncapi-knowledge-mcp==0.0.2

# or as a plugin (researcher agent + MCP server)
/plugin marketplace add derberg/asyncapi-knowledge-mcp
/plugin install asyncapi-knowledge@askasyncapi-ai

Clone & run this chat locally

git clone https://github.com/derberg/asyncapi-knowledge-mcp
./scripts/run-local.sh

Examples pin a specific version on purpose — check PyPI for the current release and always pin the one you picked. Never rely on an implicit “latest”: unpinned installs are a supply-chain risk.

Try asking:

Demo instance on limited credits (~$6/month, and rate-limited to a few questions per minute, plus a rate-limited Hugging Face deploy) — and this chat runs a small, cheap model. For day-to-day use run it locally: free, private, and you get far better answers from the stronger agents and models you already have (Claude Code, Cursor, or any MCP client).

How it’s built

One knowledge base, three ways to consume it — all open source, rebuilt weekly from the official docs.

Sources

Official AsyncAPI content

Markdown docs from the asyncapi/website repo plus the AsyncAPI 3.1.0 JSON Schema — fetched, chunked and embedded by OpenCrane into a vector index.

This chat

Vercel

A static page + two serverless functions: vector search over the knowledge base, and an agent — the gpt-5-nano model via the Vercel AI Gateway — that answers with citations back to the docs.

Run it yourself

PyPI · `uvx asyncapi-knowledge-mcp`

opencrane pack bundles the prebuilt index into a pip-installable MCP server. Runs entirely on your machine over stdio — no keys, no remote calls.

Hosted MCP

Hugging Face Space

The same server as a public HTTP MCP endpoint in a Docker Space, searching with a local embedding model — also free of API keys.