Directory of MCP Servers
187 points
14 days ago
| 22 comments
| github.com
| HN
maxwellg
14 days ago
[-]
I can't wait for first-party remote MCP servers to become more common. Right now we're taking a strange detour of everyone trying to proxy everyone else's APIs and do manual API Key juggling because platforms aren't running their own MCP servers and clients don't support the latest OAuth changes.

In a year from now, Github will run a single public Github MCP server that you will connect to via OAuth - you won't need to install it locally or faff around with tokens or environment variables at all.

reply
niel
13 days ago
[-]
> In a year from now

You can get a taste of this already.

While they still call it a prototype/beta, Sentry's MCP server [0] is a model for others to follow when it comes to convenience and usefulness.

Remote-first with OAuth. The biggest hurdle to using it as-is at the moment, is that most clients don't natively support OAuth yet, so often you'll rely on a local proxy server, like mcp-remote [1], to handle auth. Clients will catch up.

[0] https://mcp.sentry.dev/

[1] https://github.com/geelen/mcp-remote

reply
cyberge99
12 days ago
[-]
The step after that is paid access to any apis.
reply
joshstrange
13 days ago
[-]
I agree that we will probably move to first-party remote MCP servers in the near future which puts a lot of registries/etc in limbo.

That said, I think there might be a market for MCP servers that do more than the first-party client, it will really depend on what first-party support looks like. Did they implement all of their existing API in MCP or just a few parts?

However, my experience with MCP servers so far (and it’s super early days, I know), has taught me that in a lot of cases it’s better/easier to write your own MCP server/tools. A lot of MCP servers out there are sloppy and/or hard to run/debug. Since most tools are a thin layer over existing API/SDK calls it’s not hard to write (or LLM generate) the needed code which has the added bonus of giving you full control.

Even when an MCP server works 100% and is easy to run, it doesn’t always map 1:1 with the API and so I’ve run into “Yes, you can retrieve data object X but you can’t filter by Y because they didn’t implement that filter in the tool call”.

reply
meander_water
14 days ago
[-]
This is kind of what smithery does already. You can choose to install a local server, or connect to a remotely hosted server on smithery after authenticating through your GitHub OAuth.
reply
rvz
13 days ago
[-]
> I can't wait for first-party remote MCP servers to become more common

> In a year from now, Github will run a single public Github MCP server that you will connect to via OAuth - you won't need to install it locally or faff around with tokens or environment variables at all.

That sounds horrific. GitHub is known for their unreliability and centralizing everything to GitHub which isn't a good idea.

Combining two bad standards (MCP and OAuth) doesn't make remote MCP servers secure either.

reply
jjfoooo4
14 days ago
[-]
I’ve been seeing MCP compared to extensions in web browsers. Which I find telling, since I wouldn’t exactly say web extensions have been a great success - it’s a pretty niche dev market, and the security posture remains pretty anxiety inducing
reply
owebmaster
13 days ago
[-]
Extensions were a huge success, it was what made Firefox dethrone IE and then Chrome taking the lead. But then the smartphone era came and most people access the internet through them and extensions are not 1st class citizens in mobile.
reply
reustle
14 days ago
[-]
Here are a few more:

- https://smithery.ai/

- https://github.com/wong2/awesome-mcp-servers

- http://mcp.so/servers

- https://cursor.directory/mcp

But as mentioned above, there is an ongoing discussion for the Anthropic registry https://github.com/modelcontextprotocol/registry

reply
tkellogg
14 days ago
[-]
FYI https://mcp.so/ is the exact same thing as was posted. Not sure why they directed to the github instead of the actual site..
reply
Maxious
14 days ago
[-]
There's some movement on https://github.com/modelcontextprotocol/registry

> The MCP Registry service provides a centralized repository for MCP server entries. It allows discovery and management of various MCP implementations with their associated metadata, configurations, and capabilities.

reply
connor4312
14 days ago
[-]
@ VS Code we've been collaborating on this and plan to ship initial support for registries in our next release.
reply
SafeDusk
13 days ago
[-]
Instead of connecting to a server with 1000(s) of tools, I'm going the opposite direction and claim that you only need <10 sharp tools/small function for most use cases.

As an example, today I re-implemented Google's AlphaEvolve with <7 tools (https://toolkami.com/alphaevolve-toolkami-style/).

reply
joshstrange
13 days ago
[-]
100% this.

Next steps are auto-generate or auto-mashup tools (a couple of projects are doing this) and small, reusable agents that only have access to the handful of tools they need.

“Auto-mashup” refers to (I just made it up) a concept of chaining existing tools with a bit of logic so that instead of having to round trip to the LLM for common cases you can call “Get the load, and the last N log lines, and procstat the top 10 procs, …” all into a “check_server_status”. Similar to some systems that let the LLM write and reuse tools, this would be the same thing, just leveraging other/existing MCP tools. Maybe “auto-composition” is a better name.

reply
jappgar
14 days ago
[-]
Is this like 10 years ago when you could find a Directory of GraphQL Servers?

Seems silly in retrospect no?

reply
Too
14 days ago
[-]
The difference is that GraphQL requires explicit integration with every single API. With MCP you just add the endpoint (close your eyes for security issues) and voilá, several new capabilities were added that you can talk to using human language.
reply
malablaster
14 days ago
[-]
I agree. There’s no need to centralize this list.
reply
slimslenders
14 days ago
[-]
Community MCP servers available as Docker images are also being listed here https://hub.docker.com/catalogs/mcp
reply
jillesvangurp
13 days ago
[-]
There's a big market opportunity here. Countless SAAS solutions are currently trying to figure out how to deal with this new AI thing. If it has some kind of API, creating an MCP server for it isn't technically hard. You can probably generate one with an LLM. It's so easy that you wonder why this is a thing at all. Let the LLMs sort it out; this is low level plumbing stuff it shouldn't require my brain to work hard.

What is hard is integrating across SAAS solutions that haven't done this yet in a way that is secure and easy. Most MCP things out there are so far about exposing things that have a very low value. All the high value stuff is locked up behind APIs, authorization, secure networking (i.e. not publicly accessible typically), etc.

Bridging that stuff is going to generate a lot of work in the next few years and more importantly, companies are going to spend large amounts of money on this because it can deliver a lot of value to them.

People that believe that this is going to be a done deal in six months are dreaming. It's more like ten years. But that just means that there is good money to be made by people that can do this stuff and that can navigate the decades of byzantine digital cruft in the corporate world. You can already see the usual suspects (big consultancy companies) sniffing around this topic. There will be lots of such companies doing a brisk business by the end of this year.

reply
buremba
13 days ago
[-]
> People that believe that this is going to be a done deal in six months are dreaming. It's more like ten years.

You might be underestimating how fast the current ETL / integration companies can pivot to provide reliable MCP servers as the lift is pretty small.

reply
constantinum
14 days ago
[-]
[Noob doubt]

Am I getting this right? Based on the architecture/flow diagram of MCP, every SaaS app out there can build an MCP server. But you'll need a "MCP host" to make it work, right? Right now, I'm only seeing a handful of hosts — Claude Desktop and Windsurf. Who will be building these "hosts"? I'm only seeing use cases revolving around these hosts. Is there any real-life production use-cases? How will this pan out?

reply
evalstate
13 days ago
[-]
The list here https://modelcontextprotocol.io/clients has a number of Host applications, frameworks etc.
reply
happyopossum
14 days ago
[-]
Today there are a handful of client-side options, cline, Claude desktop, windsurf, Google’s ADK, etc. keep in mind though, we’re talking about a spec that was released around last Thanksgiving. It’s been like 7 months, and the pace of development has been blistering.

Once the authN/authZ stuff is fully codified and baked, we’ll see first part MCP gateways and the ability to connect to those tools with the Chatbot of your choice.

Consider what we see now as a developer preview…

reply
CSMastermind
14 days ago
[-]
I wonder if there's a market for someone figuring out how to build monetization into MCP or something similar.

Being able to offer a helpful API to the world and just getting paid whenever someone uses it would be really nice.

At the moment you have to process the payment "yourself" (even if you use a third party for that), issue an API key, etc.

reply
tomjen3
14 days ago
[-]
The market for MCP servers is the same as the market for rest endpoints: its a delivery mechanism for the underlying service.

I don't think you can make money on them, they are too simple to clone, but you can make money charging for the API. If you have a per usage license making an MCP is a very obvious choice - if you charge per seat it is mostly a question of how how sticky you are versus the competition.

reply
OtherShrezzing
13 days ago
[-]
I see value in a pay-per-execution model. I run a service which has a lot of proprietary data. Right now, if Anthropic/OpenAI wanted to use that data in their responses, they need to find me, setup an account, plug it into their chatbot, and return the data to an end user.

With some kind of MCP tip jar, they could extract the data they need and pay $0.02 for the service.

It would remove a lot of friction in the system, and could generate revenues for content & data creators.

reply
tomjen3
13 days ago
[-]
Thats exactly why I think it is such a good idea.
reply
meander_water
14 days ago
[-]
I reckon the target market would have to be non-developers (because MCP servers are easily reproducible with LLMs, they even encourage it in the docs), and you wouldn't even mention MCP. Just have a list of tools which you can optionally enable in the chat client
reply
Nedomas
14 days ago
[-]
We've built a version of this on steroids - not only a registry, but also one-click mcp hosting. Would love you eyeballs if you're into mcp: https://supermachine.ai
reply
mooreds
14 days ago
[-]
Also interesting was mintlify's decision to start one and then shut it down.

https://mintlify.com/blog/why-we-sunsetted-mcpt

Nice story of startup focus.

reply
swyx
14 days ago
[-]
> Messages flooded in from developers both within and outside our customer base, all eager to submit their servers to get listed. The validation was clear – there was significant demand for what we'd built.

I know Han and he's a smart guy but this is very very wrong lol. there's significant SUPPLY for what he built. because everyone is just trying to self promote by putting mcp wrappers of their stuff out. the hard part is the demand.

(and also the fact that anthropic is putting up an official registry so it'll be steamrolled)

reply
Yiling-J
13 days ago
[-]
I think Hugging Face will soon add an MCP category to their homepage, similar to what modelscope has done(the Chinese equivalent of Hugging Face): https://www.modelscope.cn/mcp
reply
joshwarwick15
13 days ago
[-]
reply
schappim
14 days ago
[-]
So far, I’ve catalogued over 6,000 MCP servers.

If you’re interested in the next layer beyond just discovering MCP servers, I’ve been working on https://ninja.ai — an app store for AI assistants to connect to tools via MCP, without needing to touch the command line. Think one-click installs for pipes that let agents actually do things like triage email or book Ubers.

Would love feedback if you’re experimenting in this space too!

reply
tomjen3
14 days ago
[-]
Cool. Just a heads up, ninja is also a name for a c++ build system.
reply
cadamsdotcom
14 days ago
[-]
There's a huge gap in this market for someone who can take these and make them trustworthy. Maybe the OpenRouter of MCP.
reply
tough
14 days ago
[-]
The underlying issue is always relying on a third party, on openrouter you're trusting the end model provider to not do funny business

can't really fix this

reply
cadamsdotcom
14 days ago
[-]
If you're paying said third party it's a decent mitigation.
reply
tough
14 days ago
[-]
maybe said third party could just run / veto most basic mcp servers so youcan run them on their server with some peace of mind

interesting

reply
cyanydeez
14 days ago
[-]
I don't see how they could ever be trust worthy without kneecaping the claimed benefits.
reply
AsmodiusVI
13 days ago
[-]
Docker is doing this.
reply
laacz
11 days ago
[-]
If you need directories for something, it just won't scale.
reply
devops000
14 days ago
[-]
What is a useful agent build with MCP?
reply
kordlessagain
14 days ago
[-]
I have an agent that creates new tools here: https://github.com/kordless/gnosis-evolve. I use it with Claude Desktop for a lot of different things, including browsing or searching for content, with the various crawlers that are out now. There's a crawl4ai tool that is pretty useful.
reply
ilteris
13 days ago
[-]
Couldn't you do this without mcp? Could you help me understand the value?
reply
kordlessagain
6 days ago
[-]
Yes, absolutely. Can be done without it and fairly easy to even make it call remotely. The value is a framework that a client can use to make calls, however. And if I want to use Claude Desktop, much easier to use MCP to deal with the interconnect. Consider VSCode integration there is much there for me to learn. Here it's easy to use the MCP stuff but it certainly isn't necessary in a server deployment, for example. Although you may want to write a MCP server for your endpoints, if you make money off that stuff.
reply
avandekleut
13 days ago
[-]
Ive used smithery.ai as well.
reply
867-5309
14 days ago
[-]
Model Context Protocol
reply
asdev
14 days ago
[-]
You don't need MCP you just need function calling
reply
hughdbrown
14 days ago
[-]
Yeah, but there is a distinct advantage to using a standard.

Suppose you want your agent to use postgres or git or even file modification. You write your code to use MCP and your backend is already available. It's code you don't have to write.

reply
jappgar
14 days ago
[-]
Are we still writing code?
reply
revskill
13 days ago
[-]
We write to fix the bullshits from ai.
reply
mindwok
14 days ago
[-]
Yes because we should all be building function calling implementations for the same 10 SaaS services rather than using 10 standard MCP servers.
reply
ukuina
14 days ago
[-]
But the standard servers should be hosted by the service provider, like mcp.slack.com as a counterpart to api.slack.com

Why should I be self-hosting ANY local MCP server for accessing an external service?

reply
reustle
14 days ago
[-]
That is being done as a stop gap until official servers are released. Ideally you are writing a server for your own product/service, or custom local work.

i.e. I wrote a server for water.gov to pull the river height prediction nearby for the next 24hr. This helps the campground welcome message writing tool craft a better welcome message.

Sure that could be a plain tool call, but why not make it portable into any AI service.

reply
cyberge99
12 days ago
[-]
Can you share this?
reply
Too
14 days ago
[-]
Remote MCP servers can do prompt injection that instruct your local agent to do something else other than only the expected tool call. https://embracethered.com/blog/posts/2025/model-context-prot...
reply
lyu07282
13 days ago
[-]
That flaw isn't introduced by the MCP server necessarily it can already be present in the API data it returns, you will never be able to protect yourself against someone injecting a malicious prompt that calls your code eval tool to open up a reverse shell on your MacBook Pro.
reply
owebmaster
13 days ago
[-]
that's not the case, MCP has a feature, samplings, that allow MCP servers to run prompts using the client model.
reply
lyu07282
13 days ago
[-]
Oh boy, you know at least the infosec people are going to get a good laugh from this clown show
reply
jappgar
14 days ago
[-]
I find it funny that vibers trust AI to write their entire platform but don't trust it enough to eval a curl statement.
reply
laidoffamazon
14 days ago
[-]
Is there a better “universal” or standard framework to do itv
reply
asdev
14 days ago
[-]
you don't need any universal standard, you just need functions specific to your app's use case
reply
djohnston
14 days ago
[-]
you can leverage MCPs without building any app at all.
reply
coderstartup
14 days ago
[-]
Super cool, Thanks
reply