Hey! I just launched YAMCP (pronounced “YAM-C-P”), an open-source CLI that bundles any number of Model Context Protocol (MCP) servers, local or remote, into unified local workspaces for your AI apps.
The problem yamcp intends to solve: Juggling multiple MCP servers across different AI tools means editing different config files and digging through separate log files whenever something breaks.
What it offers: YAMCP lets you create dedicated “YAM” workspaces organized however works best for your work styles, by function (e.g. coding, design, research), by app (Cursor, Claude, GitHub-Copilot), or any custom combination in between and expose them as Yet-Another-MCP. Scan an entire YAM workspace for issues, connect your AI apps to a single gateway, and manage everything from one simple gateway.
Why YAMCP?
* Organize your servers locally: Group MCP server by AI application or workflow purpose.
* Scan and Catch errors early: Scan workspaces for misconfigurations before they disrupt your work.
* Single gateway per each workspace: Point all your AI tools at one MCP instead of constantly swapping servers and manage your servers through a single cli environment locally.
* Unified management Add, remove, and update servers with a cli command.
* Centralized logging to debug all servers from one place. Track every request and response across all your MCP connections in one place.
This is very cool! I’ve used workflow builders for AI-based workflows before, but the ability to add custom code as work blocks is incredibly useful. A great idea would be to include a CLI to visualize the existing code. It might not be trivial, but current-gen AI could help extract modules from the code.
Does it also run the container in a TEE to attest to code execution, or does it only use a TPM to load the keys for signing resources? Either way, it sounds like a very cool idea! I have to dive into the code.
Thanks for the feedback. That is a problem that we noticed as well. The windows are in fact videos that are played in a loop in a picture-in-picture mode, but in the browser there is a short pause between loops. We are looking to find a workaround to see if we can make the pause less obvious or smooth it out. It was a weekend project we did last weekend while was shelterd-in-place. Definitely needs improvment.
Also a feedback. Since you’ve already put the effort to make it a website it might be worth add a filter tab/button so the users can filter them based on different dimensions(e.x city, equity, ...)
Like the Idea though I’m not a designer! Very good strategy to start your MVP with human powered engine, but I guess eventually you would need to make it AI powered to automate the search if you gain traction and need to scale. It would be too much effort to just do human powered search to just show the results while you are not sure if the users would be interested in any of them! Just to look back in history, Yahoo was a human powered search and google is an AI powered search and only one of them is alive today:)
Can come very handy for developers, specially for presentations or during training sessions. It would be great if it supports runtime engines/interpreters to support code execution for different languages. like Jupyter but with extended set of languages (e.g. javascript, perl, ruby...)
you can easily live anywhere if you are a mobile developer and work remotely from any city even outside the states. however if you are looking to work in office Seattle, NYC, Portland, DC
What is your website?! is it just a content website or you have some sort of services serving your requests in backend. Application DDos attacked can hugely impact your website if they can put load on your backend and eventually trash your servers.
Some useful solutions for those scenarios:
-proper authentication/authorization mechanism
-keep stats per user or session and set quota on the number of requests per user or session
-add captcha to your registration pages to prevent them from scripting your signup.
The problem yamcp intends to solve: Juggling multiple MCP servers across different AI tools means editing different config files and digging through separate log files whenever something breaks.
What it offers: YAMCP lets you create dedicated “YAM” workspaces organized however works best for your work styles, by function (e.g. coding, design, research), by app (Cursor, Claude, GitHub-Copilot), or any custom combination in between and expose them as Yet-Another-MCP. Scan an entire YAM workspace for issues, connect your AI apps to a single gateway, and manage everything from one simple gateway.
Why YAMCP? * Organize your servers locally: Group MCP server by AI application or workflow purpose.
* Scan and Catch errors early: Scan workspaces for misconfigurations before they disrupt your work.
* Single gateway per each workspace: Point all your AI tools at one MCP instead of constantly swapping servers and manage your servers through a single cli environment locally.
* Unified management Add, remove, and update servers with a cli command.
* Centralized logging to debug all servers from one place. Track every request and response across all your MCP connections in one place.
Check out the repo for install steps and examples: https://github.com/hamidra/yamcp
Planning to add server evaluations and an initial security scan to score MCP servers’ reliability and performance.
Would love for you to try it and share what features you’d like to see next!