Javier Redondo — Scale Venture Partners - MCP is the new WWW - June 2025
FreeMCP is the new WWW
FreeFree tier
Inputs: text
About Javier Redondo — Scale Venture Partners - MCP is the new WWW - June 2025
This article, 'MCP is the new WWW' by Javier Redondo of Scale Venture Partners, argues that the Model Context Protocol (MCP) is a foundational standard for AI agents, analogous to the web browser for humans. It explains how MCP enables LLMs to discover and access resources across the internet via a unified format, and covers adoption by major players (OpenAI, Google, Microsoft), client types (chat apps, IDEs, agent frameworks, workflow orchestrators), server-side production, and security gaps that require centralized solutions like app stores.
Key Features
Provides a standard mechanism for LLMs to discover and access resources across the internet
Unified format for capability discovery and a standard control loop for input/output
Adopted by major AI ecosystem players: Anthropic, OpenAI, Google, Microsoft
Supports multiple client types: chat apps, IDEs, agent frameworks, workflow orchestrators
Enables agentic 'supernodes' in deterministic workflow pipelines
OpenAI API now supports MCP, turning chat completions endpoint into an MCP client
Pros & Cons
Pros
- Standardizes how LLMs discover and invoke tools, reducing custom middleware
- Rapidly adopted by all major AI players since November 2024
- Works across diverse clients: chat apps, IDEs, frameworks, and orchestrators
- Allows products to easily integrate with thousands of tools through a single API change
- A common protocol eliminates the need for per-integration community effort
Cons
- Security gaps: agents may exfiltrate data due to mislabeled or confused servers
- Attackers can leverage server instructions to take undesirable actions with other tools
- MCP servers can freely change tools and instructions at any time, making safety guarantees ephemeral
- No central search or discovery mechanism for MCP servers (missing a 'Google' for MCP)
- Authentication and predictive registration of servers are still unresolved
Best For
Building AI agents that interact with external tools and servicesEnhancing coding agents with access to developer productivity and infrastructure toolsAdding integrations to agent frameworks without custom function callingCreating workflow orchestrators with agentic supernodesEnabling any application to connect to a universe of MCP servers via API parameter change
FAQ
What does MCP stand for?
Model Context Protocol, introduced by Anthropic in November 2024.
What are the security concerns with MCP?
Data exfiltration via mislabeled servers, attackers leveraging server instructions, and the risk that a server can change its tools/instructions at any time.
Which companies have adopted MCP?
OpenAI, Google, Microsoft, and others in the AI ecosystem.
What is the main limitation for finding MCP servers?
There is no central discovery or search engine for MCP servers yet; a 'Google for MCP' is missing.