MCP-Workspace

MCP-Workspace is an All-in-One MCP server that provides a complete AI development workspace in a single service. It combines file system operations, code execution in safe sandboxes, Web app deployment with domain handling, Excel/CSV data processing, and image generation capabilities (including Mermaid diagrams and visualizations) under one roof. The server is designed for multi-tenant use with session isolation, and it exposes tools such as fs_read, fs_write, fs_ops, fs_replace, fs_search for file management, exec for Python/Node.js sandbox execution, preview_frontend for one-click web deployment with wildcard domain support, and generate_image for charts and visuals. It also supports optional knowledge-base search and web crawling through kb_search, kb_read, crawl_url, and web_search when configured. Connections to AI platforms (like Dify, FastGPT, and Cherry Studio) can be established using SSE (server-sent events), enabling your agent to perform file operations, run code, deploy web apps, process data, and generate visuals seamlessly during conversations. The server enforces multi-tenancy and path safety to keep each user/session isolated within their virtual workspace. To use it, point your client or platform to the SSE endpoint (for example, http://your-server:8000/sse) and pass tenant headers such as X-User-ID and X-Chat-ID to ensure proper isolation.

Prerequisites: Docker (recommended), or a compatible Python 3.12 environment with the necessary system dependencies installed. If opting for Docker, you can use the provided docker-compose setup. If you prefer a native Python setup, ensure you have Python 3.12+ and install required Python packages from a requirements file.

Option A: Docker deployment (recommended)

1. Ensure Docker and Docker Compose are installed on your machine.
2. Clone the repository and navigate to the project directory:

git clone <repository-url>
cd <project-directory>

3. Start the services in detached mode, building images if needed:

docker-compose up -d --build

4. Monitor logs and verify the SSE endpoint is reachable:

docker-compose logs -f

5. If dependencies change, rebuild and restart as needed:

git pull
docker-compose up -d --build

Option B: Native Python (uv) setup (if you must run without Docker) Prerequisites: Python 3.12+, pip, and system dependencies required by the server (e.g., for OCR or image processing).

1. Create a virtual environment and install requirements:

python3.12 -m venv venv
source venv/bin/activate
pip install -r requirements.txt

2. Run the server directly (adjust paths as needed):

uv --directory /path/to/mcp-filesystem run run_server.py /path/to/allowed/dir1 /path/to/allowed/dir2

3. Ensure the server binds to the expected host/port and that the SSE endpoint is accessible.

Prerequisites recap: Docker is the easiest path and reduces environment mismatch. If you must run natively, ensure Python 3.12+ and all required system libraries are installed, and be prepared to manage dependencies and environment setup manually.

• The mcp-workspace server is designed for multi-tenant isolation. Always pass X-User-ID and X-Chat-ID headers from your client to ensure correct session scoping.
• For security, ensure allowed directories are explicitly configured and that path safety checks are enabled to prevent directory traversal attacks.
• The SSE-based integration is the recommended transport when connecting with AI platforms like Dify, FastGPT, or Cherry Studio. Use the provided SSE URL (e.g., http://your-server:8000/sse) and supply the tenant headers in requests.
• If you enable optional tools such as kb_search, kb_read, crawl_url, or web_search, configure their respective enabled flags (e.g., kb.enabled=true, web_crawl.enabled=true) in your environment or configuration to tighten security and control usage.
• The all-in-one design means you should monitor resource usage (CPU, memory, and I/O) per tenant to maintain performance and fairness across sessions.
• When deploying with wildcard domains, ensure your TLS termination and DNS are configured correctly to support per-session subdomains (e.g., user123_chat456.proxy.your-domain.com).