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

Key Concepts

• Emergent Intelligence: Intelligence arising from the interaction of individual, self-interested units (neurons, market participants, employees, AI agents).
• Agentic Economy: An economy where AI agents can autonomously find, interact with, and purchase services from other agents, businesses, or tools.
• Actors: Self-contained, Docker-based software units with well-defined inputs and outputs, representing a new way to ship and integrate software.
• MCP (Agent Message Communication Protocol): A standard for agentic interaction, enabling agents to communicate and discover tools.
• Tool Discovery: A key feature of MCP that allows agents to dynamically discover and use tools based on the workflow, without pre-defined configurations.
• API Marketplace: A platform (like APIFY) where developers can publish and monetize their tools (actors) and agents, making them accessible to a wider ecosystem.

Emergence of Intelligence

The speaker begins by exploring how intelligence emerges in various systems:

• Biological Systems: Neurons, initially individual cells, develop connections and learn to communicate to pursue their own interests (e.g., obtaining nutrients). The collective behavior of a large number of neurons results in what we call intelligence.
• Markets: Market behavior is more intelligent than individual participants due to the mutual interaction of self-interested members, leading to collective intelligence.
• Companies: Internal communication platforms like Slack can facilitate the emergence of company-wide intelligence, exceeding the capabilities of individual employees.

The central question is: How will general intelligence emerge in computing systems? The speaker posits that AGI will arise from the interaction of multiple AI agents pursuing their own goals, facilitated by a communication protocol like MCP.

APIFY and the Agentic Economy

• APIFY Overview: APIFY is a marketplace of 5,000 tools (actors), primarily focused on web scraping and data extraction. These actors are built by APIFY and a community of creators who monetize their tools.
• Actors as Software Units: Actors are self-contained Docker containers with defined inputs and outputs, making them easily integrable into other systems. Examples include a Google Maps scraper that extracts more data than the Google Places API.
• Integration Capabilities: APIFY provides SDKs (TypeScript, Python, OpenAPI, C#, CLI) and integrations with workflow automation tools (Make, Zapier, Clay) to facilitate actor usage.
• MCP Integration: APIFY integrates with MCP, allowing AI agents to call actors on the marketplace using an API key or OAuth.

MCP and Tool Discovery

• MCP as a Standard: MCP is presented as a leading standard for agentic interaction, surpassing alternatives like OpenAI's API and Google's A2A.
• Tool Discovery's Importance: Tool discovery is highlighted as the "killer feature" of MCP. It allows agents to dynamically discover and use tools based on the current workflow. This is crucial because manually publishing thousands of tools through OpenAPI would be impractical and increase the risk of errors.
• Dynamic Tool Selection: Clients connect to an MCP server and dynamically discover tools to use based on the workflow.

The Challenge of Autonomous Agent Interaction

• Dependency on Human Developers: Currently, agents still rely on human developers to grant them access to services (e.g., providing API tokens for Zapier apps). Agents cannot independently find and purchase services.
• The Need for Autonomous Purchasing: To achieve higher levels of intelligence, agents need the ability to find and purchase services autonomously.

Proposed Solutions for Autonomous Purchasing

1. Agent Self-Subscription: Agents could subscribe to services themselves (e.g., using email, credit cards, solving captchas). This is deemed impractical due to the need for real person verification and other complexities.
2. Central Identity and Payments Provider: A central authority (e.g., Verifier, Coinbase's X42, Stripe, Mastercard, Visa) could manage agent identities and payments. While promising, this approach faces challenges in establishing a new payment system.
3. Centralized Marketplace of MCP Services (APIFY Store): This approach involves a marketplace where developers publish tools (actors) with their own payment information. Agents can access all services with a single API token.

APIFY's Solution: A Marketplace for Agentic Services

• Developer Monetization: Developers provide their credit card and account details to third-party services, set pricing, and publish their tools on the APIFY marketplace.
• Ecosystem Expansion: This allows for rapid scaling of the ecosystem, even without direct cooperation from target services.
• Nested MCP Servers: MCP servers can be nested, allowing for the creation of a hierarchical ecosystem. A parent server can provide access to tools from nested MCP servers.
• Monetization Flow: Actors charge users, and the developers receive the money and pay for external services.

Demo: Clot for Desktop and APIFY Integration

The speaker presents a demo using Clot for Desktop, integrated with the APIFY MCP server:

1. Finding a Venue: The agent uses an actor called "rag web browser" (a Google search and data fetching tool) to find the venue of the AI Engineer World's Fair in San Francisco.
2. Scraping Twitter: The agent discovers and adds a Twitter scraping actor to its context and uses it to retrieve the latest tweet from the AI Engineer conference.
3. Using a Competitor's Service (Browserbase): The speaker demonstrates how APIFY can publish a competitor's (Browserbase) MCP server on its marketplace without their direct involvement.
4. Email Subscription: The agent uses Browserbase (via APIFY) to fill out an email subscription form on the AI Engineer website.

Scaling the Ecosystem and Monetization

• Accessibility: Anyone can publish tools or agents on the APIFY store and monetize them, gaining access to the existing AI clients and the broader ecosystem.
• Creator Revenue: APIFY paid over $4 million to its creators last month, with actors generating over $500,000 per month.
• Ecosystem Growth: The platform has 1 million monthly visitors and is focused on scaling the ecosystem.

Open Questions

• Value of Autonomous Tool Discovery: Will autonomous tool discovery provide real value, or will it be too unreliable? The speaker is optimistic that as language models improve, tool discovery will become more valuable.
• Agent Trust: How can agents trust other tools or agents?
• AGI Enablement: Can autonomous agent interaction enable AGI?

Conclusion

The speaker argues that enabling AI agents to autonomously find, interact with, and purchase services is crucial for achieving higher levels of intelligence in agentic systems. APIFY's marketplace, combined with MCP and tool discovery, offers a promising approach to building an agentic economy where agents can truly operate independently. The demo showcases the potential of this approach, while the open questions highlight the challenges that remain.