​
AI clients/Plugins

You can use MCP servers in any AI clients which support MCP.

If you haven’t used any AI client before, you can try Cherry Studio(https://www.cherry-ai.com/)

Next I will show how to configure using scanpy-mcp.

​
MCP setting

​
running in local

locate the MCP server, by which(Linux/Mac) or where(Window):

which scanpy-mcp

Refer to configure

"mcpServers": {
  "scanpy-mcp": {
    "command": "/home/test/bin/scanpy-mcp",
    "args": [
      "run"
    ]
  }
}

​
running in remote

mcp servers are running in remote server, but chatting in local AI client.

run mcp in your server by:

scanpy-mcp run --transport shttp --port 8000

Make sure you have forwarded the necessary ports. If you’re using VS Code, Trae, Cursor, or similar development environments, they will automatically handle port forwarding for you. If automatic port forwarding is not available, you’ll need to run the following command locally:

ssh -fNL 8000:localhost:8000 user@your.server.host

Then configure your MCP client in local AI client, like this:

"mcpServers": {
  "scanpy-mcp": {
    "url": "http://localhost:8000/mcp"
  }
}

​
Prompt setting

It’s important to configure a system prompt that directs the LLM to focus on scRNA-seq analysis and tool calling. Below is an example that you can adapt to your specific use case:

# scRNA-seq analysis Assistant 

You are a bioinformatician specializing in Python tools (scanpy, anndata). 

## Core Behavior
- Default to AnnData format for data analysis
- Execute tools sequentially, one at a time
- Only use parameters explicitly specified by user
- Display images using markdown format
- Convert scanpy/Seurat code to appropriate MCP tools

## Error Handling
When errors occur, troubleshoot and resolve before proceeding to next tool.

## Execution Rules
- Run only tools necessary for the specific user request
- No additional tool calls beyond what's required
- Set parameters only when specified in requirements
- Focus on completing the requested analysis efficiently

​
Configure in Trae

see it in https://youtu.be/HXPqaDvjKvg