As part of my bigger goal to make the web more agent-friendly, this weekend i decided to tackle google. The "AI-native" search APIs like Tavily and Exa exist, but they require setup and don't actually use Google's results.

So I built something simple - a proxy that takes Google search URLs and returns the results as clean markdown instead of HTML. You literally just change "google.com" to "googllm.com" in any search URL.

```bash

# Returns 500KB of HTML:

curl "https://google.com/search?q=AI+news"

# Returns clean markdown:

curl "https://googllm.com/search/AI+news"

```

*What it does:*

- Serves normal HTML to browsers (so humans can use it normally)

- Returns markdown to everything else (curl, fetch, LLM agents)

- Supports all Google search types: web, images, news, scholar, shopping, etc.

- No auth needed for testing (10 requests/hour free)

*Technical approach:*

- Content negotiation based on Accept headers

- Caches results to avoid hammering Google

- Simple pricing: 0.5¢ per search after free tier

I built this over a long weekend because I was tired of writing HTML parsers for every project. The whole thing is designed around a single principle: make Google search results consumable by LLMs without any complexity.

*Questions for HN:*

- Is this approach too simplistic? Should search APIs be more complex?

- How do you currently handle search in your LLM applications?

- Any concerns about the proxy approach vs. building from scratch?

The llms.txt documentation is intentionally comprehensive (2500 tokens) so any LLM can understand and use it immediately.

Live demo: https://googllm.com

API docs: https://googllm.com/llms.txt

Would love feedback on the approach and any edge cases I might have missed.