Hi Tip-Sheeters,

Thanks for everyone who joined the NordicAPIs LiveCast: Asynchronous APIs for AI and Data Science recently. If you're interested in skimming to catch some of the interesting bits, here are a few:

​12:45 - How asynch patterns used by agents have an impact on your API design​

​39:14 - How learning enterprise patterns is beneficial for API developers​

​49:24 - What skills should data scientists and software developers learn right now​

Thanks again to NordicAPIs -- I encourage you to sign up for the NordicAPIs Digest for good coverage of the APIs space.

Giving context and capabilities to AI agents: MCP, Skills, CLI, Code Mode

The online world has been hashing out the best way to give LLM-powered agents and coding tools access to additional sources popular options are:

​Model Context Protocol (MCP) -- an open-source standard with a lot of momentum in the last year. It involves developing a client and server and can be used locally or across the network. MCP provides tools, resources, and prompts.

​Code mode - Is a new technique that CloudFlare

​Agent Skills - An open format for giving agents instructions in natural language. The core of a skill is a `SKILL.md` file.

Command line interface (CLI) tools - These are the commands you run from a (usually linux) command prompt. They can be local commands like `grep` or API wrappers like `aws` and `az`. These weren't created for for agents, but agents can use them, especially if they're well-known CLI tools that are referenced in a lot of training data (blog posts, documentation pages, tech books).

Code Mode for MCP - To reduce the number of tokens sent to LLMs (which are the cost driver with agents), CloudFlare proposed Code Mode, which seems like a hybrid of MCP and code generation to save tokens. FastMCP supports this in version 3.1 but it's probably early days for this technique.

Here are some articles to get a taste of the discussion:

​Why CLI Tools Are Beating MCP for AI Agents - Good summary of the recent argument for CLIs instead of MCP. It has lots of examples of CLIs, such as `gh`, `kubectl`, `az`.

​MCP is Dead, Long Live MCP - Charles Chen writes a detailed explanation of the benefits of using CLI (good for local connections) and MCP (good for enterprise cases across HTTP).

​Code Mode: the better way to use MCP - CloudFlare article explaining "code mode" to build TypeScript code to run in a sandbox and make fewer calls to an LLM.

​Skills vs MCP tools for agents: when to use what - LlamaIndex wrote this explanation of both MCP and Skills, with actual examples of skills.

​MCP vs. CLI is the Wrong Fight - For those interested in testing metrics, the Smithery team ran some benchmarks of LLMs using MCP, CLIs, and raw APIs using 3 example APIs.

​CLI vs. MCP vs Code Mode - More specs, this one including Code Mode. The Portofcontext team using Stripe APIs and compared MCP, CLI and Code Mode.

​Enterprise Challenges with MCP Adoption - Christian Posta shared a talk from KubeCon EU about lessons learned using MCP.

​Filing Your Taxes by Talking to an AI - Mark Headd created a partial MCP prototype using the IRS Direct File open source code. He raised some interesting issues and suggested ideas how they might be addressed.

More Useful Resources

Here's a roundup of useful articles or posts I've read in the last month or so:

​Baseball data roles - The Pittsburgh Pirates, a U.S. Major League baseball team is hiring for data engineer, analyst, and scientist roles in their baseball side.

​Lead data scientist at Hudl - I discussed some of the Hudl/StatsBomb SDK examples in my book (read that chapter in English or Spanish) so I'm always interested to see what they're doing. This is a high-impact role they're hiring for.

​No AI without APIs - Gartner "ThinkCast" video discussing why APIs are the foundation of agentic AI.

​Arrazo UI from Jentic - Arazzo is an extension of the OpenAPI spec that allows you to design workflows that included multiple API calls. The Jentic team created a UI to visualize an Arazzo spec.

​Mastering Postgresql - I'm using Postgresql on my new anchor project, and I came across this free ebook, which has snippets from two Manning books.

​Cornell students studying AI use in baseball - As AI and robotics is added to real-world scenarios, there are many human impacts. A Cornell student is studying MLB integrating the Automated Ball-Strike System into the tradition-heavy sport of American baseball.

​Interview: How to be a 10x Engineer - Gergely Orosz had an interview with a successful software engineer about what's made them successful in their career. Some good nuggets in there.

Keep coding,

Ryan Day

👉 https://tips.handsonapibook.com/ -- no spam, just a short email every week.