Mcp Gateway and Registry
Mood
calm
Sentiment
mixed
Category
other
Key topics
The MCP Gateway and Registry project by IBM sparks discussion on the MCP ecosystem, its applications, and the challenges of implementing MCP gateways, with varying opinions on its potential and limitations.
Snapshot generated from the HN discussion
Discussion Activity
Very active discussionFirst comment
1h
Peak period
50
Day 1
Avg / period
17.7
Based on 53 loaded comments
Key moments
- 01Story posted
Aug 25, 2025 at 1:30 AM EDT
3 months ago
Step 01 - 02First comment
Aug 25, 2025 at 2:41 AM EDT
1h after posting
Step 02 - 03Peak activity
50 comments in Day 1
Hottest window of the conversation
Step 03 - 04Latest activity
Aug 27, 2025 at 2:49 AM EDT
3 months ago
Step 04
Generating AI Summary...
Analyzing up to 500 comments to identify key contributors and discussion patterns
The problem with MCP is that it cannot reliably scale, so some abstraction is helpful, i.e. separate “agent” with its own instructions and a predefined set of tools.
Assuming the latter, are there any viable non-developer MCP clients?
ChatGPT has limited MCP support as of now (read only use cases with deep research) but the expectation is that full MCP support might be dropping soon.
Writing a client capable of parsing and using that description isn’t much easier.
Unsurprisingly, it appears there’s no real spec: https://www.utcp.io/about/RFC
Fair enough, but when restricting yourself to the problem of tool calling, giving the AI the ability to call APIs in a simplified way, rather than inventing a new communication protocol like MCP, is, in a lot of cases, better.
MCP, is just a really good M2M documentation.
API's, even simple JSON data requests from websites and services were becoming increasing blocked behind cloudflare or hidden as a paid feature, MPC has revived (some of) the open data standards of the internet.
MCP allows to both: - Mount these into your chatbot of choice - Use these in any automation (custom chatbot or other LLM flow) if your framework supports MCP
Tbf I am still under the impression that for the later use case you would rather use the (HTTP) API directly because MCP doesn't allow you to customize the tool descriptions and if you truly want to make your automation perform well you need to iterate on those in my experience.
The idea is each plugin runs in their own wasm vm with limited network/file system access. Plugins are written in any language, as long as they can compile to WASM and publish to OCI registry (signed & verified with sigstore)
Recently, Microsoft released their own version of hyper-mcp named Wassete[2]
Ideally, I want to make it like a gateway with more security & governance features in this layer.
[1]: https://github.com/tuananh/hyper-mcp [2]: https://github.com/microsoft/wassette
Or do we just want CorporateSoftTM to build and own everything in just one big single commercial platform?
I would love it if loads more businesses could start up.
The romantic startup vision sounds nice, but most of those businesses would just fail after wasting capital and talent that could've been used better elsewhere. It's messy and inefficient.
>> Or do we just want CorporateSoftTM to build and own everything in just one big single commercial platform?
It's gonna get acquired anyways... what do you think the endgame of 99% of startups is? Building a company?
Burning money to reinvent things is fine with me if it trains people up, pays people a salary, creates some original ideas, leads to more choice in the market.
MySQL, MariaDB, PostgreSQL, TimescaleDB, CockroachDB, YugabyteDB, Teradata, Snowflake, TiDB, SQLite, DuckDB, and more probably.
I think it's led to great choice and a great sharing of ideas. Theres also been a lot of failures too, that's true, but I'm sure some of those failures have led to a few unique features across those products too.
So now with AI, for me it's the same process, just accelerated. Yes, there is lots of junk coming out there, but I think there's lots of good too.
Not sure which of these tools is currently the best but they are clearly needed.
I am testing MetaMCP to expose internal MCP servers to the cloud version of Claude, enabling me to use Home Assistant and some other internal services in verbal conversations with Claude app on the phone.
I think 3rd party MCP server use is available in other AI assistants as well but Claude is the only one who allows using them in Europe.
I can only wish that this would become a standard approach with any AI assistants, including the built in phone assistants like Siri, but Apple is so so far behind and always trying to squeeze every bit of money wherever they can, I doubt they will ever do that.
This approach enables better integrations with local services specific to different regions, and I would somewhat compare this to RSS feeds which were popular 15 years ago. Would be nice if most e-commerce sites would provide their own MCP servers for managing carts, placing and tracking orders, etc. I don't want to see a world where single monopoly grabs this market and makes things shit again (like it happened with social media), especially for those who live outside of USA.
This needs to be able to support plugins to override behavior. I don't want my models having a giant tool registry, I want them to ask an oracle for the best 3 options, and I'd like to be able to inject specific behavior on a per MCP basis.
Mostly I find working with MCPs pretty easy, but there is some minor benefit to hosting mcp servers that are more involved to set up. The thing that I would pay a small amount for (which is why I'm building it myself) is a MCP oracle that configures the MCP server set for an agent based on the task it's working on.
Although i believe too for development teams it might be valuable, all devs get automatically access to all services with 1-click install etc etc. But question is how unique/powerful is it.
Everyone installs this gateway as their only "MCP", then at a central location we can add different MCP tools and everyone automatically gains access to them.
And yes to teh comments about people making a million of these... so very very many of them.
I think the agentic use cases aside, the client side of MCPs are still lacking quite a bit and would need to mature to be able to catch up to the spec. I feel a lot of use cases exist outside of fully automated agentic approaches, since we can't really rely on LLMs yet to produce at a human level.
The underserved cases rely a lot on prompt and resource management at the moment. Being able to iterate and share those across teams to provide easy starting points to delegate tasks is something I feel would be workable for the current iteration of AI assisted work, outside of pure software engineering.
Hopefully other clients join VS Code Copilot to allow more varied approaches than just simple tool calls here. I think Copilot's approach on prompt and resource management isn't quite the best approach either though. It is still early days for MCPs in general so i think we'll see a lot of experiments in this space.
IBM posting as if it were a solo open-source developer.
give it a try here: https://getaiko.app/downloads
One of the bigger challenges we are currently observing is missing authorization for MCP servers.
We actually also just released a mcp-gateway[1] initially focusing on adding OAuth authorization to streamable HTTP MCP servers.
Want the full context?
Jump to the original sources
Read the primary article or dive into the live Hacker News thread when you're ready.