Show HN: An MCP server that turns async-work practices into tools

Posted by benbalter 3 days ago

More than a decade ago, I adopted the self-imposed rule, if I answer a question more than once, the third time I need to be able to answer with a URL. Today, I published one very large URL - a book distilling what I learned from helping people work remotely at GitHub, and I wanted to rethink my rule for the age of AI.

What if, instead of a URL, I could create an interactive experience that could tailor the guidance to your particular situation?

What I ended up building was an Open and Async Advisor MCP server. To install (in claude or any other AI):

> claude mcp add open-async -- npx -y @open-and-async/mcp

Transparently, yes, it's from a book I wrote which launched today, but the MCP server is open source and free to use. No purchase required. It knows the key principles of the book, and has specialized tools like `draft_decision_doc`, `convert_meeting_to_async`, `score_status_update`, `triage_sync_vs_async`.

This is an experiment for me, and I'm genuinely curious if it's helpful for others.

Comments

Comment by polotics 3 days ago

The readme is so thick with breathless LLMisms that it makes one wonder what is the point. That's a bit sad because the book's website and blog appear interesting, the author can write. Still: shall I spend 10 bucks by buying and reading the book, or by consuming tokens on LLMs that read it?

Comment by benbalter 2 days ago

> the book's website and blog appear interesting, the author can write

I'll take the compliment!

README was 100% intended for robots, not humans (book is 100% for humans, FWIW). I'd be impressed if you can spend $10 on tokens interacting with the MCP. That's one reason why I went with an MCP over a skill. A lot of the logic is offline/token free.

Comment by znhskzj 3 days ago

This has been very inspiring for a MCP design novice like me; I can now implement executable tools, reference knowledge, and user workflows in my own MCP.

Comment by benbalter 2 days ago

This was my first "real" MCP server (previously I made an MCP server to control a Furby for GitHub Universe) and I learned a lot. A few things I learned (not an expert):

* Tool vs. Resource: Does it compute or does it serve * Prompts -> Tools * Deterministic vs. Probabilistic is a choice per tool

Comment by taevirus 2 days ago

How does `triage_sync_vs_async` handle team-specific norms — fixed rubric, or does it lean on the model's judgment from context?

Comment by benbalter 2 days ago

Fixed rubric. It works on regex matching across interpersonal tension, active incident, sensitivity, and ambiguity. If a pattern matches, it weights sync over async.

Comment by mjhea0 3 days ago

So nice to see a new library written in JavaScript instead of TypeScript.

Comment by suzuridev 3 days ago

The "book as an MCP server" framing is the part that interests me. I run a small MCP server myself (semantic search over government documents), and the thing I keep noticing is that the interface changes what people ask — nobody asks a PDF "what should I do in my situation," but they'll ask a tool that. Shipping specialized tools (draft_decision_doc etc.) over one generic ask_the_book tool also seems like the right call — clearer intent per call, less context spent on tool descriptions. I noticed you went with keyword search over a summary corpus plus citations rather than embedding the full text — curious whether that was a licensing decision (keeping the manuscript out of the package) or you found summaries-with-citations just work better for this kind of prescriptive content than semantic retrieval over the full prose?

Comment by benbalter 2 days ago

Partially licensing, but also tokens + efficiency. As noted above, the book is written for humans (anecdotes, examples, etc.). Robots just need the facts and can extrapolate. If I were building a hosted service, a custom LLM with embeddings would probably beat both approaches.