atlantis

Atlantis is an MCP-hosting setup that allows you to expose and orchestrate tools and dynamic functions across a distributed network. The core idea is that you run a Python MCP host locally (the Remote) which can load dynamic functions from the dynamic_functions/ folder and install or manage third-party MCP servers on the fly. Tools and servers can be discovered, shared, and invoked by other agents or clients in the Atlantis cloud, enabling cross-machine automation and collaborative tool usage. The readme outlines how to register Atlantis as an MCP server in your local Windsurf/Cursor setup and how to connect to the Atlantis cloud, making Atlantis tools available for use by Claude-like clients or other MCP clients. You can also port existing MCP services (e.g., an Open Weather MCP) to the uvx-based runtime and wire them into the same network, increasing interoperability across languages and runtimes.

Prerequisites:\n- Python 3.8+ (preferably Python 3.13 as noted) and Node.js installed on your machine.\n- uvx (Python MCP runtime) and/or npx (Node MCP client) depending on which tools you plan to use.\n- Access to a working internet connection to fetch MCP packages.\n\nStep-by-step:\n1) Clone the repository:\n git clone https://github.com/ProjectAtlantis-dev/atlantis-mcp-server.git\n\n2) Install Python dependencies for the remote (Python MCP host):\n cd python-server\n python -m venv venv\n source venv/bin/activate # on macOS/Linux\n pip install -r requirements.txt\n\n3) Install Node.js dependencies (for the MCP client host/tools):\n npm install # or use npx as documented in the config\n\n4) Configure runServer (as described in the Quick Start):\n Edit python-server/runServer or copy to a runServerFoo script and set:\n --email=youremail@example.com\n --api-key=YOUR_API_KEY\n --host=localhost\n --port=8000\n --cloud-host=wss://projectatlantis.ai\n --cloud-port=443\n --service-name=home # or any unique remote name\n\n5) Start the Python MCP host (remote):\n python server.py \n --email=youremail@example.com \n --api-key=foobar \n --host=localhost \n --port=8000 \n --cloud-host=wss://projectatlantis.ai \n --cloud-port=443 \n --service-name=home\n\n6) To run Atlantis as an MCP server via npx (standalone):\n npx atlantis-mcp --port 8000\n\n7) Optionally add Atlantis to your Claude/Open Weather workflows or Windsurf Cursor configs following the examples in the README.\n

Tips and notes:\n- Atlantis uses a dynamic function and dynamic MCP server model; new tools can be dropped into dynamic_functions/ and loaded on start.\n- To add third-party MCP servers, place a JSON config under dynamic_servers/ with the same mcpServers structure used in examples.\n- If you encounter authentication issues when connecting to the Atlantis cloud, ensure your email and API keys are correctly set and that you’re using the same email across cloud and local remote.\n- The weather/open-weather example demonstrates porting an existing MCP to uvx format; you can reuse that approach to integrate other services.\n- When using npx or uvx, ensure the targeted ports do not conflict with other services on your machine.\n