🧠 Food for thought

Implications, context, and why it matters.

Model Context Protocol (MCP) adoption remains unclear despite rising interest

• Anthropic introduced Model Context Protocol (MCP) in November 2024 1. OpenAI uses it 2. Google DeepMind support is coming 2. Zed, Replit, and Sourcegraph back it 1. Metrics are missing.
• The protocol lacks standardized authentication (identity verification) and authorization (access control) mechanisms 3. SaaS multi-tenancy hosting remains hard 3, and security risks like prompt injection plus tool permission gaps can enable data exfiltration 2.
• Most MCP servers install locally (on a developer’s machine) 4. AWS and Microsoft Azure are working to make MCP portable and cloud-compatible 4. The ecosystem remains experimental, which falls short of Dedalus’s “5 lines of code” pitch at scale.

SaaS companies can reach agents by building MCP servers now

• For business-to-business (B2B) software vendors, turning services into MCP servers gives AI agents and developers a direct path without middlemen platforms, like how REST APIs opened programmatic access in the 2010s 3.
• Mintlify, Stainless, and Speakeasy offer tools (developer tools that turn API documentation into working servers) to produce MCP servers from docs 3. Cloudflare (a cloud and edge network provider) and Smithery (a platform for publishing and hosting MCP tools) provide hosting that eases setup 3.
• Discovery options include mcpt (a command-line discovery tool for MCP tools), Smithery (a public registry), and OpenTools marketplaces (directories for AI tools) 3. The MCP Servers directory (a listing of MCP servers) adds reach 5. Early entries gain visibility, but results hinge on developer and enterprise uptake that remains unproven.