# Pagewatch

**Category:** 💻 Developer Tools  
**Repository:** https://github.com/toolshedlabs-hash/web-access-skills  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/pagewatch

## Description
A real browser for agents. Turn any URL into clean markdown, a PNG screenshot, or a PDF.

## Claude Desktop Quick Installation
Heuristic fallback — verify the package name and runner against the repository README before running it. Uses `npx` (confidence: low):

```json
"mcpServers": {
  "pagewatch": {
    "command": "npx",
    "args": ["-y","pagewatch"]
  }
}
```

## Documentation & README

# web access skills for agents

[![skills.sh](https://skills.sh/b/toolshedlabs-hash/web-access-skills)](https://skills.sh/toolshedlabs-hash/web-access-skills)

Give your agent real web access. One install, no signup forms. The agent registers
its own key and starts on free credits, so the first call just works.

Three skills, one shared account:

- **screenshot-url** turn any URL into a PNG or JPEG. Viewport or full page.
- **url-to-pdf** save a web page, or a chunk of raw HTML, as a clean PDF.
- **url-to-markdown** read any URL and get clean markdown back, JavaScript sites included.

They all run pages in a real hosted browser, so a screenshot looks like a real
screenshot and a client-rendered site (React, Vue, and the rest) comes back with its
actual content instead of an empty shell. Most agent runtimes have no way to take a
picture of a page or print one to PDF on their own. These fill that gap in one step.

## Install

```bash
npx skills add toolshedlabs-hash/web-access-skills
```

That pulls in all three. Add one at a time if you prefer:

```bash
npx skills add toolshedlabs-hash/web-access-skills/screenshot-url
npx skills add toolshedlabs-hash/web-access-skills/url-to-pdf
npx skills add toolshedlabs-hash/web-access-skills/url-to-markdown
```

## MCP server

The same three tools (screenshot, url to pdf, web read) are also served as a hosted
MCP server, so any MCP client can call them without installing the skills. It is a
remote streamable-http endpoint, no local process to run.

```
https://pagelens.dev/mcp
```

Point your MCP client at that URL. On the first call the server mints a trial key for
the agent and returns real content, the same free-to-start flow the skills use. It is
listed on the Model Context Protocol registry as `dev.pagelens/pagewatch`.

## Use

Each skill is one bundled Python file, standard library only, no pip install. Run it
from the skill directory.

```bash
# screenshot
python3 screenshot.py "https://news.ycombinator.com" -o hn.png --full-page

# pdf
python3 url_to_pdf.py --url "https://example.com/report" -o report.pdf

# read as markdown
python3 url_to_markdown.py "https://example.com/some/article"
```

## How the key works

The first time a skill runs it registers a pagewatch key for the agent and stores it
under your config directory. No form, no dashboard, no API key to copy from anywhere.
That key starts with a small free trial, so the very first call returns real content.

When the trial runs out the skill prints a short block asking a person to confirm one
email. They open a link, enter an email, and click the button in the message we send.
That releases about 200 more free credits. It is a one time step per key, it is free,
and there is nothing to pay or sign up for.

The exit code tells the agent what happened without parsing text: `0` success, `3`
trial used up so relay the link, `4` free credits used up, `5` the request could not
complete, `6` new keys are being rate limited so wait and retry.

## Pricing

Free to start. A small trial on the first run, then about 200 free reads once a human
confirms one email. After that, in credits, a read costs 1, a screenshot 2, and a PDF
4. A credit is meant to be worth about half a cent, so 200 credits is roughly one
dollar. Nothing can be charged today. The service is in free preview and has no
payment system connected, so there is no card on file anywhere and no way to spend
real money on it yet.

## What it will not do

- It obeys robots.txt. If a site disallows a path, you get an honest refusal, not a
  workaround.
- It does not defeat bot walls, solve CAPTCHAs, or spoof a fingerprint. If a site
  refuses automated traffic, you get told that, plainly.
- It does not reach private or internal network addresses.
- Anything behind a login or a paywall stays there.
- A failed or refused call costs zero credits. You pay for successful calls only.

## Config

All optional, set as environment variables:

- `PAGEWATCH_BASE_URL` which service to talk to. Defaults to the hosted service.
- `PAGEWATCH_TOKEN` use a token directly instead of the stored one.
- `PAGEWATCH_TOKEN_FILE` where the token is kept. Defaults to a file under your user
  config directory, readable only by you.

## About

Made by toolshed, a small studio that builds tools for agents and the people who run
them. Maker pen name Cal. These skills are free and MIT licensed. They call a hosted
pagewatch service that meters usage with credits and is in free preview right now.
Only successful calls ever draw a credit. Questions or feedback: toolshedlabs@gmail.com