LoginSign Up
← Back to all sparks
A

Apify

ANALYTICS
Velocity5.0

Web-scraping platform is reshaping itself around AI agents — MCP, permissions, and OpenAPI surfaces.

web scrapingai agentsmcpactor permissionsopenapiagent infrastructure
◆Current state
Apify continues to optimize for AI-agent consumption. Recent shipments include interactive OpenAPI documentation for standby Actors with auto-attached API tokens, an approval modal for full-permission Actors (least-privileged defaults), multiple datasets per Actor for cleaner output structure, and a redesigned MCP configurator covering Claude Desktop, Claude.ai, Claude Code, Antigravity, Cursor, ChatGPT, Codex, and VS Code. The mcpc universal MCP CLI client and Dynamic Actor memory rounded out the prior month.
◆Where it's heading
Apify is converging on a single thesis: be the scraping and Actor execution infrastructure that AI agents call into. Every recent release either improves how agents discover and run Actors (MCP configurator, OpenAPI Endpoints tab, mcpc CLI) or hardens what happens when they do (full-permission approvals, dataset structure, dynamic memory). The product is no longer marketing itself primarily as scraping — it's marketing itself as agent-callable web automation.
◆Prediction
Expect tighter cost-attribution and audit trails for agent-initiated runs, more nuanced permission scopes, and continued expansion of supported MCP-aware client editors. Standby Actors as a deployment model are likely to see more first-class support — they're a natural fit for agent-callable APIs.

◆Recent moves

  1. 16d ago

    Interactive OpenAPI documentation for standby Actors

    An Endpoints tab with interactive OpenAPI docs for standby Actors that auto-attaches the API token — a meaningful DX upgrade for any developer or agent author trying to figure out how to call an Actor.

    View source ↗
  2. 17d ago

    Full-permission Actors now require approval

    A one-time approval modal on full-permission Actors continues the least-privileged push and is especially relevant for AI-agent invocation, where users may not realize what an Actor can access.

    View source ↗
  3. 28d ago

    Multiple datasets for Actors

    Actors can now output to multiple datasets each with its own schema. Useful for separating successful results from failures (Website Content Crawler now does this for failed URLs) and producing cleaner structured data overall.

  4. 28d ago

    Deploy agents faster with an improved MCP configurator

    MCP configurator refresh shortens setup across an explicit list of client editors. Reads as a deliberate doubling-down on the agent-callable-Actor positioning.

  5. 28d ago

    MCP configurator refresh (republish)

    Republished MCP configurator announcement — same content as the dated entry.

    View source ↗
  6. 28d ago

    Multiple datasets for Actors (republish)

    Republished version of the multi-dataset announcement on a different feed — same content as the dated entry.

    View source ↗