jesse_portal

thanks! i'll go back and try to re-do some of the text. do you mean the landing page or the module descriptions specifically? all the content is links to reputable sources so i'm assuming those are fine.

very good point there. establishing the target audience and pre-requisite knowledge is pretty important.

Made this free site that teaches everything required to get a job as either inference engineer or for model training by analyzing a bunch of current job requirements for job postings by top labs.

I'm working through it on my own but would love any advice or feedback.

Yeah 100%. For the reputation we'll have rating. Instead of rating per server/gpu though like vast.ai and clore.ai it will be for the provider, and it will match bond/credit ratings, e.g., AAA, B, C. Providers start at C (which is way cheaper but with a huge risk). As the provider begins to move up their 'credit score' increases and their contracts appreciate to match.

Regarding "verified" and "secure cloud", the idea is to have "terms". Each contract can list requirements. For example, we have secure/enterprise gpu owners who are required to collect KYC for anyone using their gpus. This can be specified as a "term" on the contract.

For the gpu owner:

list an open 4090 contract, anyone can redeem:

gpubook provider sell 4090-standard-v1 --date 2026-06-01 --ask 8.40 --terms open

list a secure/enterprise contract that requires buyer identity + compliance info:

gpubook provider sell h100-premium-v1 --date 2026-06-01 --ask 42.00 --terms identity,organization,compliance,workload

For the gpu buyer:

only buy from AAA-rated providers, and only if the terms are open:

gpubook buy h100-premium-v1 --date 2026-06-01 --bid 39.00 --min-rating AAA --open-only

buy from at least A-rated providers, and accept enterprise/KYC terms:

gpubook buy h100-premium-v1 --date 2026-06-01 --bid 42.00 --min-rating A --accept-requirements identity,organization,compliance,workload

So the book is still one market for h100-premium-v1, but each order carries its terms. A bid only matches an ask if the price crosses, the provider rating is high enough for the buyer, and the buyer has accepted the contract’s requirements. Open contracts and enterprise/KYC contracts can sit on the same book, but buyers can filter or restrict what they’re willing to take.

I rent and manage GPUs on sites like vast.ai and clore.ai. The prices are all over the place. You will see 5090s from $0.30/hr to $5/hr, plus bandwidth which is set by the gpu owner and can range from $0.4 per TB up to $12.00 per TB.

This is a concept for what an order book would look like for GPU compute instead of the current listing model. It would use standardized SKUs so users know what they are getting. Its pre product, I'm curious to know what people think.

Hey Hacker News,

I built a simple registry to make it easier to find and share remote MCP servers: https://remote-mcp-servers.com

I know there are a lot of MCP server registries, but most of them don't focus on remote access so I've created this one.

If you run a remote enabled MCP server, please add yours!

Feedback welcome on features and usability. Thanks!

Have been implementing MCP resource support in my app.

Took the approach of letting users add resources directly to their current message which adds a snapshot of the resource to the message history.

Also added the ability to 'pin' a resource to a chat which adds it automatically to the top of the chat history (experimented with adding it to the bottom but then the agent would comment on it directly).

I'm hoping it enables some cool use cases.

Hi HN,

Jesse from Portal One here. We all know AI agents get lost or struggle with complex, multi-step tasks.

Our article explores a way to make agents more reliable using dynamic Model Context Protocol (MCP) servers. Instead of just being a static API wrapper, the MCP server actively guides the agent by changing the available tools and information based on the current task state, an adaptive environment.

To illustrate this "guided tour" concept, we built a very simple demo server where an agent plays a Number Guessing Game.

The game itself is trivial, but the mechanics demonstrate the core idea:

- Lobby state: Agent sees only `start_game` tool.

- Game started: `start_game` is removed; `guess_number` and `give_up` tools appear.

- Guess made: The `guess_number` tool's own input schema adapts (e.g., guess must now be > 50).

- Game ends: Tools revert to the lobby state.

The full article details this flow and how the server uses MCP notifications (`toolListChanged`, etc.) to keep the agent updated.

The real importance isn't the game, but what this dynamic guidance concept unlocks for helping agents stay on task.

GitHub Repo & Demos:

https://github.com/portal-labs-infrastructure/number-guessin...

Sharing our experience implementing our production MCP server using OAuth, TypeScript, and the new ProxyOAuthServerProvider class from the MCP SDK.

Check out the companion GitHub repo: https://github.com/portal-labs-infrastructure/mcp-server-blo...

Yeah, I was also going to recommend litellm. I've been using it for my LLM assistant app and was able to add support for pretty much any model or provider endpoint by switching out one line of code.