# MCP tools

Connect with any MCP client (see the [Quickstart](https://skryp.dev/docs/quickstart)). Your agent sees these tools. Leave settings on auto unless the task needs them fixed; every tool's result states how it was fetched and what it cost.

| Tool | What it does | Main arguments |
| --- | --- | --- |
| `scrape` | Read one or up to 50 known URLs | `url` / `urls`, `formats`, `focus`, `country`, `strict_country`, `render`, `actions`, `max_credits`, `login` |
| `read` | Read more of a stored long page without fetching it again (free) | `ref`, `section`, `focus`, `offset` |
| `map` | List a site's URLs from sitemaps and links | `url`, `search`, `limit` |
| `search` | Web results from several engines, optionally from a country | `query`, `limit`, `country`, `scrape_results` |
| `extract` | Typed fields by schema or prompt, checked against the page | `url`, `schema`, `prompt`, `country`, `login` |
| `inspect` | The page's own data calls and whether they replay without a browser | `url`, `country` |
| `recipe` | Save, run and test repeatable collection methods | `action`, `id`, `spec`, `params` |
| `crawl` | Durable background crawl | `url`, `limit`, `max_depth`, `include_paths`, `exclude_paths` |
| `job` | Status, results, cancel or resume of crawls | `action`, `id` |
| `collect` | Records with the fields you name from a listing page | `url`, `fields`, `limit`, `max_pages` |
| `dataset` | Collections that stay current: create, refresh, records, changes, export, schedule | `action`, `id`, `url`, `fields` |
| `monitor` | Scheduled checks of a page, extraction or recipe with diffs | `action`, `kind`, `target`, `interval_s`, `webhook`, `credit_cap` |
| `compare` | One page from several countries with converted prices | `countries`, `url`, `schema`, `base_currency` |
| `browser` | A live browser session: open, snapshot, click, type, read, close | `action`, `session`, `url`, `ref`, `text`, `login` |
| `archive` | Pages as they were, change timelines, diffs, field history, old URLs | `action`, `url`, `at`, `since`, `until`, `schema` |
| `workflow` | Saved multi-step jobs (filters, pagination, downloads), run signed in to a saved login: save, run, status, results, schedule, cancel, resume | `action`, `id`, `definition`, `inputs`, `login` |
| `account` | Balance, usage, available routes and saved logins | — |

The same list with every tool's input schema is at [/mcp-tools.json](https://skryp.dev/mcp-tools.json), for clients and agents that read it directly.

## Habits that save credits

- Use `focus` to receive only the sections about your question.
- Set `max_credits` to cap what a page may cost: `1` keeps it on our own network, `10` allows residential proxies but not advanced unblocking (25).
- Call `account` before a large job: it reports the plan, the credits left and the prices.
- Big pages return a `ref`: call `read` for more instead of scraping again.
- Turn repeated collections into a `recipe` or a `dataset`; refreshes then need no browser and no model.

## dataset

`dataset` keeps a listing current. `create` takes a page and the fields you want; `refresh` reports added, changed and removed records by stable ID. A refresh that fails or stops early never marks records as removed.

## Signed-in pages: login and workflow

A saved login is your own account on a site, connected once in the dashboard (Logins): you sign in through a live browser, and Skryp keeps only that site's cookies and storage, sealed for your workspace. Pass its id or name as `login` to `scrape`, `extract` or `browser` to read pages as that account.

`workflow` repeats the clicks, filters, pagination and downloads you would do by hand, with inputs (for example one run item per keyword), limits and an optional schedule. Each item is checkpointed, so an interrupted run resumes without repeating finished items.

If the site signs the account out, the result says `needs_login`: nothing from the sign-in page is passed off as data or charged, and a workflow run pauses until you reconnect the login, then continues where it stopped.

Source: https://skryp.dev/docs/mcp