Agent Integration

The Parallax agent doesn't just use this documentation as a one-time training input — it can query it live at inference time via MCP (Model Context Protocol) and Context7.

How it works

Your prompt → Parallax Agent
                    ↓
          Context7 MCP Server
                    ↓
       docs.thriv.es/llms.txt    ← index of all pages
                    ↓
       Relevant page content     ← fetched on demand
                    ↓
          Grounded response

When you prompt the agent and your question touches a topic covered in these docs (Love2D patterns, API usage, best practices), Context7 fetches the relevant page and injects it into the agent's context window before generating a response. This means:

• Docs improvements immediately improve agent answers — no retraining required
• The agent cites sources when it draws from documentation
• Adding new articles (patterns, guides, recipes) directly makes the agent smarter

Setting up in Claude / Cursor

Add the Context7 MCP server to your AI assistant:

Claude (via MCP config)

{
  "mcpServers": {
    "context7": {
      "command": "npx",
      "args": ["-y", "@upstash/context7-mcp@latest"]
    }
  }
}

Then in your prompts, append use context7 or reference a library:

Add a screen shake effect to my Love2D game. use context7

Context7 will look up the Parallax docs (and the official Love2D docs) and feed the relevant content to the agent.

Cursor

Add the same MCP config to .cursor/mcp.json in your project root.

The /llms.txt endpoint

This site exposes an index at /llms.txt — a plain-text map of all documentation URLs and descriptions, following the llms.txt standard. Context7 and other MCP tools use this to know what's available before fetching.

You can also point any LLM tool at it directly:

https://docs.thriv.es/llms.txt

Writing docs that help the agent

Since articles feed directly into agent context, write them with the agent in mind:

• Be specific — "use dt for movement" is more useful than "handle timing correctly"
• Include code examples — the agent learns patterns from examples, not abstractions
• Name patterns explicitly — "the entity table pattern", "the bump.lua collision pattern" — named patterns are easier for the agent to recognise and apply
• Keep pages focused — one topic per page means more precise retrieval

Every new article you add to the game-dev/ section makes the agent a better Love2D collaborator.