# Capturing web pages inside Harvey: a custom MCP connector for screenshots and PDFs

> Source: <https://dev.to/dennis-ddev/capturing-web-pages-inside-harvey-a-custom-mcp-connector-for-screenshots-and-pdfs-1l5f>
> Published: 2026-11-09 21:14:29+00:00

*Written by the SnapRender team. SnapRender is the capture service used in this guide; the setup works on its free plan.*

A lot of legal work starts with a web page. A competitor's advertising claim, a platform's terms of service on the day a client signed up, a listing that infringes a trademark, a privacy policy that changed last month. The usual routine is to open the page, take screenshots by hand, print to PDF, rename the files and attach them to the matter, then go back to Harvey to ask questions about what the page says.

Harvey can now call external tools through custom MCP connectors. If one of those tools captures web pages, Harvey can open a URL, keep the capture as a file in the conversation, and work with the page text directly. This guide sets that up and shows the prompts that make it useful.

With SnapRender connected, Harvey can:

Cookie banners are removed before the capture, so it shows the content rather than a consent dialog.

Harvey's custom connectors talk to remote MCP servers over HTTPS with OAuth. Harvey registers itself with the server automatically, so there is no client ID or secret to copy.

```
   Name:            SnapRender
   Server URL:      https://app.snap-render.com/mcp
   Authentication:  OAuth
   Client ID:       (leave empty)
   Client secret:   (leave empty)
   Scope:           (leave empty)
```

If you do not see the option to add custom connectors, a Harvey workspace admin needs to enable them for you.

Click **Connect**, sign in to SnapRender and click **Allow**. A free account is enough to start.

In Harvey Assistant, turn SnapRender on under **Sources**.

Harvey keeps files that connectors return, but it does not display inline images from tool results. If you simply ask for "a screenshot", the image is dropped and you get a description at best.

So ask for the capture **as a file**:

Take a full-page screenshot of [https://example.com/terms](https://example.com/terms) as a PDF file.

SnapRender then returns the capture as an attached file, and it stays in the conversation like any document you uploaded.

**Terms or policy on a given day.**

Take a full-page screenshot of [https://example.com/terms](https://example.com/terms) as a PDF file and note today's date. Then summarize the limitation of liability clause.

**Advertising claims, desktop and mobile.**

Capture [https://example.com/product](https://example.com/product) on desktop and on iPhone as files, then list every performance or pricing claim on the page and where it appears.

**Reading without capturing.**

Read [https://example.com/privacy](https://example.com/privacy) as text and list what it says about data retention and transfers outside the EU.

**Comparing versions.** Capture a page today and again next week, then ask Harvey to compare the two files. For pages you need to watch continuously, SnapRender can also capture on a schedule and alert you when a page changes, set up from its dashboard.

It is worth being precise about the limits, especially for evidence work.

The free plan includes 200 captures a month. Every screenshot or page read uses one capture, and failed captures are not counted. Paid plans start at $9 a month for higher volume.

The full setup page, with an FAQ, is here: [SnapRender for Harvey](https://snap-render.com/ai-connectors/harvey).
