AI agents

Coding agents can read these docs as Markdown and search them through an MCP server, so they write Thunderclap code from the current docs rather than from memory. Nothing here needs an account.

Connect the docs MCP server

The server lives at https://thunderclap.dev/mcp (Streamable HTTP, no sign-in). It has three read-only tools: list_docs lists the pages, search_docs finds the sections that match a question, and read_doc returns a page in full.

With Claude Code:

claude mcp add --transport http thunderclap-docs https://thunderclap.dev/mcp

With Cursor, in your project or in ~/.cursor/mcp.json:

{
  "mcpServers": {
    "thunderclap-docs": { "url": "https://thunderclap.dev/mcp" }
  }
}

With VS Code:

{
  "servers": {
    "thunderclap-docs": { "type": "http", "url": "https://thunderclap.dev/mcp" }
  }
}

Other clients take the same address: add it as a remote (HTTP) MCP server.

Read the docs as Markdown

Every page of these docs has a Markdown version at the same address with .md added, such as /docs/quickstart.md. For agents without MCP:

• /llms.txt lists every page with a one-line summary, following llmstxt.org.
• /llms-full.txt holds every page in one file, to paste into a model's context.

What to tell your agent

Agents do best with the two rules most often got wrong: the browser uses @thunderclapjs/client with the public key only, and publishing and tokens happen on your server with @thunderclapjs/server and the secret. A line in your project's agent instructions (AGENTS.md, CLAUDE.md) is enough:

Realtime uses Thunderclap: look things up with the thunderclap-docs MCP server.
Browsers use @thunderclapjs/client with THUNDERCLAP_KEY only; the backend publishes
and signs tokens with @thunderclapjs/server and THUNDERCLAP_SECRET.