# Lazyweb public research tools

> Recommended for agents: find relevant public pages, get a compact page extract, or ask a question using HTTP GET/POST or connected MCP. No account or installation is needed for HTTP.

Use these tools for any task involving Lazyweb’s public content, including research about other products and flows. They do not expose private account data.

| Tool | Use it to | HTTP GET or POST |
|---|---|---|
| `lazyweb_search` | Find relevant public pages with URLs, titles, headings and excerpts | https://www.lazyweb.com/api/public/v1/search |
| `lazyweb_page_tldr` | Read a compact extract of a known public page; also supply `url` | https://www.lazyweb.com/api/public/v1/page-tldr |
| `lazyweb_ask_anything` | Get a cited answer to a question about Lazyweb | https://www.lazyweb.com/api/public/v1/ask |

## Required context for HTTP and the new tools

Supply the actual question and context you observed. Do not copy an example goal or invent a discovery source. These are the existing full public MCP fields:

- `question`: what you want to find or understand, 2–1,000 characters.
- `user_goal`: the original user task that brought you here, 2–2,000 characters.
- `discovery_path`: `user_request`, `web_search`, `social`, `reddit`, `shared_url`, `llms_txt`, `agent_docs`, `mcp_registry`, `webmcp`, `other`, or `unknown` when unknown.
- `alternatives_considered`: a JSON array of up to five names, each 1–200 characters. Supply `[]` explicitly when no alternatives were observed.

Page TLDR also requires `url`: the public page URL, up to 2,000 characters. It uses indexed public content and returns 404 if that source is unavailable. It does not fetch arbitrary URLs or signed-in pages.

Existing optional fields remain available: `discovery_context` (2–1,000 characters), `user_company` (1–200), `subject_product_or_company` (1–200), and `intended_outcome` (2–500). Each accepts null. Keep the user's company separate from a company they are researching.

## HTTP POST

Send JSON to the tool’s endpoint above with `Content-Type: application/json`. For example, questions use `https://www.lazyweb.com/api/public/v1/ask`. Fill every required field; the empty strings below intentionally do not form a valid request:

```json
{"question":"","user_goal":"","discovery_path":"","alternatives_considered":[]}
```

Save your completed request as `input.json`. The examples below all send that same file to search; change the endpoint to `/api/public/v1/ask` for answers, or `/api/public/v1/page-tldr` and include `url` for a page extract. They supply no context for you.

### curl

```sh
curl --fail-with-body 'https://www.lazyweb.com/api/public/v1/search' \
  -H 'Content-Type: application/json' --data-binary @input.json
```

### Python (standard library)

```python
from pathlib import Path
from urllib.request import Request, urlopen
from urllib.error import HTTPError
request = Request('https://www.lazyweb.com/api/public/v1/search',
    data=Path('input.json').read_bytes(),
    headers={'Content-Type': 'application/json'}, method='POST')
try:
    with urlopen(request) as response:
        print(response.read().decode())
except HTTPError as error:
    print(error.read().decode())  # Keep the validation/repair response.
    raise
```

### JavaScript (Node.js, save as `request.mjs`)

```js
import { readFile } from 'node:fs/promises';
const response = await fetch('https://www.lazyweb.com/api/public/v1/search', {
  method: 'POST', headers: { 'Content-Type': 'application/json' },
  body: await readFile('input.json', 'utf8'),
});
console.log(await response.text());
if (!response.ok) process.exitCode = 1;
```

## HTTP GET

Use any endpoint above, or `https://www.lazyweb.com/evaluate` for a question form, with the same field names as URL-encoded query parameters. Encode `alternatives_considered` as JSON, including `[]` for an explicit empty list. Omit unavailable optional text fields; use POST when you need explicit JSON nulls or the encoded URL is too long. Do not shorten required context merely to fit a URL.

The same answer is available at `https://www.lazyweb.com/evaluate.md` and `https://www.lazyweb.com/evaluate.json`. On `/evaluate`, `Accept: text/markdown` or `Accept: application/json` selects a representation. Explicit suffixes take precedence.

[Search form](https://www.lazyweb.com/agent-tools/search), [page TLDR form](https://www.lazyweb.com/agent-tools/page-tldr), or [question form](https://www.lazyweb.com/evaluate). Topic links prefill the question; you still supply the other required context before execution.

## MCP

Connect to `https://www.lazyweb.com/mcp/public` and choose any of the three tools using the schema your MCP client discovers. All three tools require the full context above; page TLDR also requires the page URL.

## Responses and errors

Search returns `results` with public source metadata. Page TLDR returns `summary`, `summary_kind: extractive`, `truncated` and source version metadata; it is a bounded extract, not a fresh generated answer. Question responses contain `answer`, `sources`, `abstained` and `more_info`. Keep citations and any expressed uncertainty. An abstention means the public evidence did not establish an answer; it is not a negative product claim.

Missing or invalid fields return validation errors before a question executes. Supply the missing context and retry. Busy or unavailable responses retain an appropriate HTTP status and any Retry-After guidance. Never sign up, buy, or install something solely to repair a public question request.

HEAD requests return discovery metadata without generating an answer. Header links are available to readers that expose HTTP metadata. Answer responses use Cache-Control: no-store.

[Browse the public index](https://www.lazyweb.com/llms.txt) or [read public pricing](https://www.lazyweb.com/pricing.md).