The rise of the agentic economy on the shoulders of MCP — Jan Curn, Apify

Emergence of Intelligence in Systems 00:01

  • Intelligence emerges from interconnected individual units (like neurons) pursuing their own interests and forming communication pathways.
  • Similar collective intelligence emerges in markets and companies via interactions among participants.
  • In computing systems, general intelligence may arise not from larger models, but through interaction and collaboration among multiple independent agents.

Introduction to Apify and the Actor Marketplace 02:45

  • Apify is a marketplace hosting about 5,000 tools called "actors," which are self-contained Docker-based software units with well-defined input and output.
  • The majority of actors are web scraping or data extraction tools, but there are also tools for data processing and workflow automation.
  • The marketplace allows integration via SDKs for various languages and offers easy access through CLI and popular workflow platforms.
  • Community creators can build and monetize actors, creating a software marketplace dynamic.

Role of MCP in Agentic Interaction and Tool Discovery 05:04

  • MCP integration enables AI agents and workflows to easily call any actor on Apify's marketplace using a single API key or account.
  • A key feature is "tool discovery," allowing agents to dynamically find and utilize the right tool for their workflow, overcoming context limits of OpenAPI.
  • MCP's adoption is increasing rapidly and is emerging as the dominant standard for agentic system interoperability.

Challenges in Agent Autonomy and Service Access 07:27

  • Despite available tools, agents can't fully autonomously discover and use most services, as access typically requires human-provided API tokens or accounts.
  • Human intervention remains necessary for connecting agents to third-party services, limiting agents' potential to act independently.

Approaches to Enabling Fully Autonomous Agents 09:04

  • Letting agents subscribe and pay for services themselves is impractical due to identity and payment requirements often tied to real individuals.
  • Central identity and payment providers (e.g., Coinbase, Stripe, MasterCard, Visa) are developing standards, but broad adoption will be slow due to marketplace challenges.
  • Apify proposes a centralized marketplace approach: developers provide payment and identity info once, then their services can be monetized and accessed by all agents via a single API/account, bypassing direct third-party integration.

Demonstration of Agentic Tool Usage and Integration 11:47

  • Practical examples show AI agents using Apify actors via tool discovery to answer questions, scrape Twitter, or interact with third-party services (e.g., Browserbase) potentially without the latter's awareness.
  • Nested MCP servers allow tools from multiple sources to be accessible through the marketplace, facilitating ecosystem growth.
  • The marketplace model enables rapid scaling as new tools or MCP servers can be added without explicit cooperation from original service providers.

Monetization, Ecosystem Growth, and Open Questions 15:45

  • Anyone can publish and monetize tools on the Apify store, gaining access to a large and fast-growing developer and user ecosystem.
  • More than $4 million was paid to creators in the last month, with over half a million dollars generated monthly and 1 million monthly visitors.
  • The real-world value and reliability of autonomous tool discovery remain open questions, but improvements in language models are expected to address these concerns.
  • Trust among agents and between tools is a major unresolved challenge, as is the broader question of whether agentic interaction can lead to artificial general intelligence (AGI).
  • The presenter encourages experimentation and participation in the ecosystem.