Supports Docker-based MCP server deployment and configuration as mentioned in the future roadmap for containerized server management
Enables setup and configuration through Git-based installation process for local MCP server deployment
Built with Python runtime support for local server execution and MCP server discovery functionality
1 MCP Server: A MCP server that picks and configures MCP servers for you
We aim at providing only this MCP server. Then you can leave all the rest (searching servers, selecting servers, configuring servers, etc) all to this MCP server.
No need to run setup commands, no need to acquire api keys. Just need to modify one file.
Demo video: https://youtu.be/Kv2HgD9hRZ8
Set up Instruction
Simple remote setup: integration with Cursor and Claude (Option 1)
Add the following to curson or claude MCP config file.
For Cursor: Open ~/.cursor/mcp.json
For Claude: Open
- macOS:
~/Library/Application\ Support/Claude/claude_desktop_config.json
- Windows:
%APPDATA%\Claude\claude_desktop_config.json
Add the following to the file:
If you are already using other servers, the json file should look like this
(Option 2) Local Setup with STDIO
Unfortunately, up to the time this md file is updated, claude only allows stdio. So you'd have to modify server.py
to use STDIO. Find the main block in server.py
, comment out the "Streamable HTTP server BLOCK" and uncomment the "Standard I/O server BLOCK". Final code should look like this
The mcp.json should look like this:
Architecture
There are two types of search tools: quick search and a deep search.
Quick Search
When the user has an explicit goal of what type of MCP they want ("I want a MCP server that handles payment"), this tool just gives back a list of mcp servers.
Deep Search *
When the user has a high level or complex description of the goal ("Build me a website that analyzes other websites"). The LLM need to break it down into multiple steps and components (I need to analyze the website traffic, I need to analyze the website tech stack, I need to show some web data, ...), then find MCP servers for each step. If a corresponding MCP server doesn't exist, inform the user to see if we should ignore this component, break it down further, or implement it ourselves.
I refer to this as horizontal expansion and vertical expansion. Horizontal expansion is for finding independent components, vertical expansion is for finding steps that have to be done sequentially (more like fetch, analyze, generate graph). In the above example, those are all horizontal expansions.
There are multiple stages in the deep search:
- Planning stage:
- setup mcp servers:
- get and configure API keys as needed, provide users with instructions of obtaining API keys
- modify the mcp.json files.
- setup mcp servers:
- Testing stage:
- test to see if they servers are working. Call
test_server_template_code
tool, which return a simple client testing code example.
- test to see if they servers are working. Call
- Acting stage:
- build the workflow/application by calling the MCP servers
*We're supposed to put deep search as a prompt, but both cursor and claude rarely calls prompts.
Change Log:
- July 31 2025: Upgrade to 0.2.0. Added agentic planning. For complex tasks, the server now prompts the LLM to perform multi-step MCP server query.
Future
- improve the demo videos: new domain name, actual example, voice explanation
- Call For MCCP (Model Context Communication Protocol): Standard way of communicating between MCP servers. Motivation: Allow directly sending requests to other mcp servers (each mcp server might also have dependencies). (But would also need stricter supervision)
- shouldn't call functions with a leading prefix
internal_
unless instructed by MCP servers - Better database for MCP servers. It should be in structure: server, description, url, config json, (optionally, additional setup, docker, api_key, etc)
This repo is based on these repos. Huge thanks to the author and contributors of these repos.
- wong2/awesome-mcp-servers
- metorial/mcp-containers
- punkpeye/awesome-mcp-servers
Trouble shooting
- If using venv, ModuleNotFoundError even after installing the module -> delete venv and create a new venv.
This server cannot be installed
hybrid server
The server is able to function both locally and remotely, depending on the configuration or use case.
MCP of MCPs. Automatic discovery and configure MCP servers on your local machine. Integration with Claude and Cursor.
Related MCP Servers
- AsecurityAlicenseAqualityAn MCP server to run commands.Last updated -2946183TypeScriptMIT License
- AsecurityFlicenseAqualityA MCP Server used to collect MCP Servers over the internet.Last updated -318Python
- AsecurityAlicenseAqualityEasily find MCP servers using our MCP registry. Search with natural language.Last updated -15JavaScriptMIT License
- -securityAlicense-qualityAn MCP server that connects any MCP client (like Claude or Cursor) with the browser using browser-use, allowing clients to utilize existing LLMs without requiring additional API keys.Last updated -74PythonApache 2.0