AgentRender ships demo API with MCP tools for agent-driven screenshots, PDFs and page extraction

What launched

A developer posting on dev.to under the handle corbinvachal has announced AgentRender, a public demo API aimed at AI agents. The service takes a public URL and returns a screenshot, a PDF or a structured extract of the page. Its stated differentiator is first-class support for MCP, the Model Context Protocol that agent frameworks increasingly use for tool calls, meaning an agent can invoke the rendering endpoints directly rather than through hand-written integration code.

The announcement is unusually frank about how early the project is. The author describes the offering as a demo, notes that the published Stripe checkout links run in test mode only, and states outright that there is no traction to report. The purpose of the post is to find design partners and gather waitlist feedback on whether the proposed pricing is something people would actually pay.

What the endpoints do

According to the post, four routes are live: POST /v1/screenshot, POST /v1/pdf and POST /v1/extract, plus a GET /healthz status check. Screenshots come from a real Chromium browser driven by Playwright. That detail matters for agent workloads, because full browser rendering captures content produced by JavaScript, which lighter fetch-based pipelines routinely miss. The PDF and extract routes cover the cases where an agent needs a shareable document or structured data rather than an image.

The positioning is explicitly aimed at agent builders: the author contrasts AgentRender with screenshot APIs built for marketing use cases, and points to MCP integration and per-render records, described in the post as render receipts, as the core features for machine consumers.

Pricing, with an asterisk

The post lays out four tiers:

• Free: $0 for 150 renders per month
• Starter: $17 for 2,000 renders per month
• Growth: $79 for 10,000 renders per month
• Scale: $259 for 50,000 renders per month

The asterisk is billing. Every Stripe link in the announcement is in test mode, so no live charges are possible; interested users are directed to a waitlist form to register interest in free keys. The author also puts the pricing question directly to readers, asking whether $17 or $79 per month is the right anchor.

Why it matters

Rendering as a service is a well-populated category, but most of it was designed for human developers embedding thumbnails or generating reports. Agent workflows have different needs: an agent verifying a claim wants a faithful capture of what a page actually rendered, plus evidence that the capture took place. An API that speaks MCP natively removes a layer of plumbing for anyone building browsing, scraping or research agents, and per-render records fit the emerging pattern of agents needing receipts for what they observed on the web.

The caveats are real, including a solo launch, test-mode billing and no stated uptime commitments, so this reads as a design-partner opportunity rather than a production dependency. Even so, the launch is a useful signal of where API products are heading: MCP support is shifting from differentiator to expectation, and packaging web access for machines rather than humans looks increasingly like the demand curve to chase.