🤖MCP Evolves, LLMs Directly Call APIs
LLMs are cutting out the middleman with APIs
TL;DR
LLMs now bypass Model Context Protocol (MCP) servers, directly calling APIs. This shift reduces context bloat and simplifies setup, but raises concerns about API documentation and response formats.
LLMs are now bypassing the Model Context Protocol (MCP) servers to directly call APIs, marking a significant shift in how AI interacts with external services. Developers and teams using MCP will see reduced context bloat and simplified setups, but must now ensure their APIs are well-documented and optimized for machine-readable responses. This change impacts workflows involving API-heavy services and documentation sites, which need to adapt to new content negotiation standards like Accept: text/markdown for agent-oriented data.
Key Points
MCP was released in Nov 2024 by Anthropic, evolving under their stewardship until 2025.
LLMs can now execute code, reason about large codebases, and call APIs directly, reducing context bloat.
Platforms like Composio, MintMCP, and Pipedream offer generic search/execute patterns to manage credentials and tools.
Cloudflare's Code Mode uses LLMs to compose API calls into scripts for sandbox execution.
LLMs use --help command to discover CLIs, reducing reliance on MCP servers for API access.
Why It Matters
If you're using LLMs with API-heavy services, ensure your APIs are well-documented and optimized for machine-readable responses. Cloudflare's Code Mode exemplifies this shift, allowing LLMs to compose API calls into scripts for sandbox execution. This change simplifies setups but demands better API documentation and response formats.
Comments
Be the first to comment
Enjoyed this article?
Get it daily. 7am. Free. Reads in 5 minutes.
Join 3,484 builders reading daily.