SerpApi home
Register
Built for LLMs & AI Agents

Web search results, optimized for LLMs

Serve any SerpApi search result as clean markdown. Same data as JSON, zero configuration, roughly half the tokens. Works with every API.

Get started for free
JSON 
{
  "organic_results": [
    {
      "position": 1,
      "title": "Coffee - Wikipedia",
      "link": "https://en.wikipedia.org/wiki/Coffee",
      "displayed_link": "en.wikipedia.org › wiki › Coffee",
      "snippet": "Coffee is a beverage brewed from...",
      "snippet_highlighted_words": ["Coffee"],
      "favicon": "https://cdn.example.com/favicon...",
      "source": "Wikipedia"
    },
    {
      "position": 2,
      "title": "Starbucks Coffee Company",
      "link": "https://www.starbucks.com/",
      ...
    }
  ]
}
Markdown 
## Organic Results

| Position | Title | Snippet | Source |
| --- | --- | --- | --- |
| 1 | [Coffee - Wikipedia](https://en.wikipedia.org/wiki/Coffee) | Coffee is a beverage brewed from... | Wikipedia |
| 2 | [Starbucks Coffee Company](https://www.starbucks.com/) | Explore the menu, sign up for... | Starbucks |

Same data, better format

Markdown tables carry the same information as nested JSON, with fewer tokens and no brace-counting for your model.

~50%
Average token savings vs JSON
100+
APIs supported
0
Configuration required

Three ways to access

Pick whichever fits your stack, all three return the same output.

Query parameter
output=md
Route extension
/search.md
Accept header
Accept: text/markdown 
# All three produce the same markdown output:

curl "https://serpapi.com/search?engine=google&q=coffee&api_key=YOUR_KEY&output=md"
curl "https://serpapi.com/search.md?engine=google&q=coffee&api_key=YOUR_KEY"
curl -H "Accept: text/markdown" "https://serpapi.com/search?engine=google&q=coffee&api_key=YOUR_KEY"

Token savings by API

Cut token usage by 50% on average. Maximize your efficiency with massive savings of more than 90% on APIs with complex nested JSON.

API JSON Markdown Savings
Google Search 150K 39K
74%
Google Shopping 361K 38K
90%
Google Hotels 202K 22K
89%
Apple Maps 494K 43K
91%
Amazon 191K 97K
49%
DuckDuckGo 32K 12K
64%
Bing Shopping 214K 103K
52%
Walmart 106K 68K
36%

Drop-in for any agent workflow

Markdown is how modern LLMs prefer to consume context. Drop it into any agent pipeline.

AI Agent

Runs a search

SerpApi

output=md

LLM Context

50% fewer tokens

Response

Faster, cheaper

Why markdown?

Every design choice serves one goal: fit more useful data in your model's context window.

Token efficient

Half the tokens of equivalent JSON, same information.

Native markdown links

URLs rendered as proper markdown links, [title](url) instead of separate fields

Clean & readable

Natural reading flow with tables, links, and headers instead of nested braces.

No noise

Tracking fields, opaque tokens, and duplicated values stripped automatically.

Universal

Works across all 100+ SerpApi APIs with zero configuration.

Structured output

Result arrays render as clean markdown tables (organic results, shopping, etc.).

Complete search results in clean markdown

YAML frontmatter, result tables, markdown links and images, pagination — all generated automatically for every API.

https://serpapi.com/search.md?engine=google&q=coffee 
---
engine: google
q: coffee
location: Austin, Texas, United States
total_results: 1,240,000,000
time_taken_displayed: 0.42
---

# Search Results for "coffee"

## Knowledge Graph
**Coffee** — Beverage
Coffee is a beverage brewed from roasted coffee beans. Darkly colored, bitter, and slightly acidic...

## Organic Results

| # | Title | Snippet | Source |
| --- | --- | --- | --- |
| 1 | [Coffee - Wikipedia](https://en.wikipedia.org/wiki/Coffee) | Coffee is a beverage brewed from roasted coffee beans... | Wikipedia |
| 2 | [Blue Bottle Coffee](https://bluebottlecoffee.com) | Fresh roasted specialty coffee, delivered to your door. | Blue Bottle |
| 3 | [Starbucks Coffee Company](https://www.starbucks.com) | Explore the menu, sign up for Starbucks Rewards... | Starbucks |

## Related Searches
- [best coffee near me](https://serpapi.com/search?q=best+coffee+near+me)
- [coffee shops](https://serpapi.com/search?q=coffee+shops)
- [types of coffee beans](https://serpapi.com/search?q=types+of+coffee+beans)

## Pagination
- [Next page →](https://serpapi.com/search.md?engine=google&q=coffee&start=10)

Cut tokens further with JSON Restrictor

Pair markdown with the json_restrictor parameter to keep only the sections your agent uses.

https://serpapi.com/search.md?engine=google&q=coffee&json_restrictor=organic_results 
## Organic Results

| # | Title | Snippet | Source |
| --- | --- | --- | --- |
| 1 | [Coffee - Wikipedia](https://en.wikipedia.org/wiki/Coffee) | Coffee is a beverage brewed from roasted coffee beans... | Wikipedia |
| 2 | [Blue Bottle Coffee](https://bluebottlecoffee.com) | Fresh roasted specialty coffee, delivered to your door. | Blue Bottle |
| 3 | [Starbucks Coffee Company](https://www.starbucks.com) | Explore the menu, sign up for Starbucks Rewards... | Starbucks |

Free Plan · 250 searches / month

Get started

FAQs

Does this cost extra?
No. Markdown output is included at no extra cost on all plans. A search with output=md costs the same as output=json. One search credit, same as any JSON search.
Does it work with all APIs?
Yes. Markdown output works across all 100+ SerpApi APIs — Google, Bing, YouTube, Amazon, Walmart, DuckDuckGo, and every other API we support. No configuration needed.
What data is included vs excluded?
All informational content is preserved: titles, snippets, links, ratings, prices, etc. Internal tracking fields (opaque tokens, redirect URLs, duplicate numeric extractions) are stripped to save tokens.
Can I still use JSON when I need it?
Absolutely. Markdown and JSON serve different purposes. Use output=md when feeding results to an LLM, and output=json when you need programmatic access with exact field extraction.
Why not just send the JSON straight to the LLM?
Markdown carries the same information in ~50% fewer tokens. That means lower LLM costs, faster inference, and more of the context window free for your actual task.