Content retrieval

The fetch endpoint extracts clean, structured content from a web page. You get the page title, description, full text, and metadata — without writing a scraper. It returns JSON by default, or LLM-ready markdown with format=context.

curl "https://api.andiai.com/api/v1/fetch?url=https://example.com/article" \
  -H "x-api-key: YOUR_API_KEY"

Parameters

Parameter	Type	Default	Description
`url`	string	required	Absolute URL to fetch (http or https)
`format`	string	`json`	`json` or `context` (markdown with YAML frontmatter)
`effort`	string	—	How thoroughly to retrieve the page: `low`, `medium`, `high`, `max`. Omit for the adaptive default. See using the effort parameter.
`maxContentLength`	integer	200000	Maximum content length in characters. You can lower this, not raise it.
`query`	string	—	When set, the response includes `query_extracts` and `query_snippet` scoped to this query.

JSON response

{
  "url": "https://example.com/article",
  "title": "Understanding Retrieval-Augmented Generation",
  "description": "RAG combines retrieval with generation for grounded LLM responses.",
  "content": "Retrieval-Augmented Generation (RAG) is a technique that...",
  "markdown": "# Understanding Retrieval-Augmented Generation\n\nRAG is a technique...",
  "extracts": [
    "RAG retrieves relevant documents from a knowledge base before generating a response.",
    "This grounds the model's output in factual, up-to-date information."
  ],
  "snippet": "RAG combines retrieval with generation for grounded LLM responses.",
  "author": "Jane Smith",
  "date": "2026-06-15",
  "site_name": "Example Blog",
  "image": "https://example.com/images/rag-diagram.png",
  "lang": "en",
  "word_count": 1850,
  "truncated": false,
  "metrics": {
    "duration_ms": 1240,
    "cost_dollars": 0.001
  }
}

Fields like author, date, image, and site_name appear when the page provides them.

Context format

With format=context, the response is markdown with YAML frontmatter — ready to pass into an LLM context window:

curl "https://api.andiai.com/api/v1/fetch?url=https://example.com/article&format=context" \
  -H "x-api-key: YOUR_API_KEY"

---
title: Understanding Retrieval-Augmented Generation
url: https://example.com/article
description: RAG combines retrieval with generation for grounded LLM responses.
author: Jane Smith
date_published: 2026-06-15
source: example.com
lang: en
word_count: 1850
cost_dollars: 0.001
retrieved_at: 2026-07-14T21:30:00Z
---

Retrieval-Augmented Generation (RAG) is a technique that...

Using the `query` parameter

Pass a query to get passage-level extracts scoped to a specific question:

curl "https://api.andiai.com/api/v1/fetch?url=https://example.com/article&query=how+does+RAG+reduce+hallucination" \
  -H "x-api-key: YOUR_API_KEY"

The response includes query_extracts (passages relevant to the query), query_snippet (a query-focused summary), and query_hash.

Using the `effort` parameter

Pass effort to control how thoroughly the page is retrieved — the same tier names used by search modes:

curl "https://api.andiai.com/api/v1/fetch?url=https://example.com/article&effort=high" \
  -H "x-api-key: YOUR_API_KEY"

low favors speed; max spends the most time extracting content, useful for pages that are slow to load or render content client-side. Omit effort to use the server’s adaptive default. Invalid values return a 400 listing the valid tiers.

Fetch failures (422, 503) are never billed, regardless of effort.

Partial responses

A 200 response can include partial content when the page is slow to fully retrieve:

Field	Type	Description
`partial`	boolean	`true` when the returned content is a best-effort excerpt, not the full page
`retry_after_seconds`	integer	Seconds to wait before re-requesting the full document

Treat partial: true as incomplete — re-request the same URL after retry_after_seconds to get the full content. This differs from a 503, where no content was extracted at all.

Error handling

The fetch endpoint uses two distinct error codes for page-level failures:

Status	Meaning	What to do
503	Page is still being retrieved, or a transient failure occurred	Retry after the `Retry-After` header value
422	Page was reached but content could not be extracted	Do not retry — the page is not extractable

Both 422 and 503 responses are not billed. You only pay for successful extractions. Other errors (400 for a missing or invalid URL, 401, 402, 429) follow the same patterns as the search endpoint — see error handling.

Pricing

Fetch calls are billed at a flat

0.001 base per request, plus

0.05 per 1M tokens of extracted content. Failed fetches (422 and 503) are free. The billed amount is returned in metrics.cost_dollars (JSON) or cost_dollars in the frontmatter (format=context).

Search extracts as an alternative

If you already have search results and want passage-level text without fetching each page, use extracts=true on the search endpoint instead. This returns text extracts inline on each result at no additional cost:

curl "https://api.andiai.com/api/v1/search?q=how+does+RAG+work&extracts=true" \
  -H "x-api-key: YOUR_API_KEY"

The fetch endpoint is for when you need the full page content, or when you want to retrieve a specific URL that may not appear in search results.

MCP tool

The andi_fetch_url MCP tool wraps this endpoint. See Build with AI agents for setup instructions.

Next steps

Search modes

Automatic effort by default, manual control when you want it.

Query parameters

Full parameter reference including extracts.

Response format

JSON and context format output structure.

Build with AI agents

MCP server and agent integration.

Getting started

Search

Features

Resources

Parameters

JSON response

Context format

Using the `query` parameter

Using the `effort` parameter

Partial responses

Error handling

Pricing

Search extracts as an alternative

MCP tool

Next steps

Search modes

Query parameters

Response format

Build with AI agents

​Parameters

​JSON response

​Context format

​Using the query parameter

​Using the effort parameter

​Partial responses

​Error handling

​Pricing

​Search extracts as an alternative

​MCP tool

​Next steps

Search modes

Query parameters

Response format

Build with AI agents

Parameters

JSON response

Context format

Using the `query` parameter

Using the `effort` parameter

Partial responses

Error handling

Pricing

Search extracts as an alternative

MCP tool

Next steps