docs/CONTINUE_DEV_SETUP.md

Continue.dev MCP Integration Setup Guide

1. Local Development Setup (Recommended)

Continue.dev Configuration

Edit your Continue.dev configuration file: Location: ~/.continue/config.json

{
  "models": [
    {
      "title": "Claude 3.5 Sonnet",
      "provider": "anthropic",
      "model": "claude-3-5-sonnet-20241022",
      "apiKey": "your-anthropic-api-key"
    }
  ],
  "experimental": {
    "modelContextProtocolServers": [
      {
        "name": "openproblems-spatial",
        "transport": {
          "type": "stdio",
          "command": "python",
          "args": ["-m", "mcp_server.main"],
          "cwd": "/path/to/your/SpatialAI_MCP"
        }
      }
    ]
  },
  "docs": [
    {
      "title": "Nextflow Documentation",
      "startUrl": "https://www.nextflow.io/docs/latest/"
    },
    {
      "title": "Viash Documentation",
      "startUrl": "https://viash.io/docs/"
    },
    {
      "title": "OpenProblems GitHub",
      "startUrl": "https://github.com/openproblems-bio/openproblems-v2"
    },
    {
      "title": "Spatial Transcriptomics Methods",
      "startUrl": "https://github.com/openproblems-bio/task_spatial_decomposition"
    }
  ]
}

Important Configuration Notes

1. Replace the path: Change /path/to/your/SpatialAI_MCP to your actual project directory
2. Python environment: Ensure the python command points to the environment where you installed the MCP server
3. Working directory: The cwd field ensures the MCP server runs from the correct directory

Verification Steps

# 1. Navigate to your project directory
cd /path/to/your/SpatialAI_MCP

# 2. Verify your MCP server works
python -m mcp_server.main

# 3. Test CLI tools
openproblems-mcp info
openproblems-mcp tool check_environment

# 4. Generate documentation cache
openproblems-mcp download-docs

2. Alternative Setup Methods

Method A: Virtual Environment Activation

If you're using conda/virtualenv, specify the full Python path:

{
  "experimental": {
    "modelContextProtocolServers": [
      {
        "name": "openproblems-spatial",
        "transport": {
          "type": "stdio",
          "command": "/home/obi/miniforge3/bin/python",
          "args": ["-m", "mcp_server.main"],
          "cwd": "/home/obi/SpatialAI_MCP"
        }
      }
    ]
  }
}

Method B: Using Shell Script Wrapper

Create a wrapper script for more control:

File: scripts/start_mcp_server.sh

#!/bin/bash
cd /path/to/your/SpatialAI_MCP
source activate your-conda-env  # if using conda
exec python -m mcp_server.main

Continue.dev config:

{
  "experimental": {
    "modelContextProtocolServers": [
      {
        "name": "openproblems-spatial",
        "transport": {
          "type": "stdio",
          "command": "/path/to/your/SpatialAI_MCP/scripts/start_mcp_server.sh"
        }
      }
    ]
  }
}

3. Remote Deployment Options

Option A: HTTP Server (Future Enhancement)

Our current MCP server uses stdio transport. To deploy remotely, you'd need an HTTP wrapper:

# Future: http_server.py
from fastapi import FastAPI
from mcp_server.main import handle_call_tool, handle_list_tools

app = FastAPI()

@app.post("/mcp/call-tool")
async def call_tool_endpoint(request: dict):
    result = await handle_call_tool(request["name"], request["arguments"])
    return {"result": [item.text for item in result]}

Option B: SSH Tunnel (Current Solution)

For remote access with current stdio transport:

# On remote server
ssh -R 8022:localhost:22 remote-server

# Continue.dev config for SSH tunnel
{
  "experimental": {
    "modelContextProtocolServers": [
      {
        "name": "openproblems-spatial",
        "transport": {
          "type": "stdio",
          "command": "ssh",
          "args": [
            "remote-server",
            "cd /path/to/SpatialAI_MCP && python -m mcp_server.main"
          ]
        }
      }
    ]
  }
}

4. Testing Your Integration

Step 1: Test MCP Server Standalone

cd /path/to/your/SpatialAI_MCP

# Test tools
openproblems-mcp tool echo_test message="Hello MCP"
openproblems-mcp tool check_environment

# Test resources
openproblems-mcp info

Step 2: Test Continue.dev Integration

1. Restart VS Code after updating config
2. Open Continue.dev sidebar (Cmd/Ctrl + L)
3. Ask a spatial transcriptomics question:

"Help me create a Nextflow pipeline for spatial transcriptomics quality control"

4. Verify MCP tools are available - the agent should:
   • Check your environment with check_environment
   • Access our documentation resources
   • Create files using write_file
   • Validate pipelines with validate_nextflow_config

Step 3: Debug Connection Issues

Check Continue.dev logs:

• Open VS Code Developer Tools (Help > Toggle Developer Tools)
• Look for MCP connection errors in Console

Common issues:

# Issue: Python not found
# Solution: Use full Python path
"command": "/usr/bin/python3"

# Issue: Module not found
# Solution: Check working directory and installation
"cwd": "/correct/path/to/SpatialAI_MCP"

# Issue: Permission denied
# Solution: Make script executable
chmod +x scripts/start_mcp_server.sh

5. Production Deployment Architecture

graph TD
    A["👩‍💻 Computational Biologist<br/>Continue.dev in VS Code"] --> B["🔗 MCP Connection<br/>(stdio transport)"]

    B --> C["📡 OpenProblems MCP Server<br/>python -m mcp_server.main"]

    C --> D["🛠️ Local Tools"]
    C --> E["📚 Cached Documentation"]
    C --> F["🐳 Local Docker"]
    C --> G["⚡ Nextflow Workflows"]

    H["🌐 Continue.dev Docs Crawler"] --> I["📖 Real-time Documentation"]
    I --> J["Nextflow.io"]
    I --> K["Viash.io"]
    I --> L["GitHub OpenProblems"]

    A --> H

6. Example Agent Interactions

Scenario 1: Environment Setup

User: "Check if my environment is ready for spatial analysis"

Agent Response:

I'll check your spatial transcriptomics environment using the OpenProblems MCP server.

[Uses MCP tool: check_environment]

Results:
✅ Docker: Available (version 28.1.1)
❌ Nextflow: Not found
❌ Viash: Not found
❌ Java: Not found

I recommend installing the missing tools:
- Nextflow: curl -s https://get.nextflow.io | bash
- Viash: curl -fsSL get.viash.io | bash
- Java: sudo apt install openjdk-17-jre-headless

Scenario 2: Pipeline Creation

User: "Create a spatial QC pipeline following OpenProblems standards"

Agent Response:

I'll create an optimized spatial transcriptomics QC pipeline using the latest
Nextflow best practices and OpenProblems templates.

[Uses MCP resources and tools to]:
1. Get Nextflow DSL2 best practices
2. Access spatial workflow templates
3. Create optimized pipeline file
4. Validate syntax and configuration

[Creates file: spatial_qc_pipeline.nf with production-ready workflow]

7. Troubleshooting Common Issues

MCP Server Not Starting

# Check if server starts manually
cd /path/to/your/SpatialAI_MCP
python -m mcp_server.main

# If it fails, check:
1. Python environment has required packages
2. Working directory is correct
3. No import errors in the logs

Continue.dev Not Detecting MCP Tools

# Verify MCP protocol compliance
openproblems-mcp info

# Check Continue.dev logs in VS Code Developer Tools
# Look for MCP connection status messages

Tools Failing to Execute

# Test tools individually
openproblems-mcp tool list_directory directory_path="."
openproblems-mcp tool validate_nextflow_config pipeline_path="test.nf"

# Check file permissions and paths
ls -la /path/to/your/SpatialAI_MCP

8. Advanced Configuration

Resource Limits

{
  "experimental": {
    "modelContextProtocolServers": [
      {
        "name": "openproblems-spatial",
        "transport": {
          "type": "stdio",
          "command": "python",
          "args": ["-m", "mcp_server.main"],
          "cwd": "/path/to/your/SpatialAI_MCP"
        },
        "timeout": 30000,
        "maxConcurrentRequests": 10
      }
    ]
  }
}

Multiple MCP Servers

{
  "experimental": {
    "modelContextProtocolServers": [
      {
        "name": "openproblems-spatial",
        "transport": {
          "type": "stdio",
          "command": "python",
          "args": ["-m", "mcp_server.main"],
          "cwd": "/path/to/your/SpatialAI_MCP"
        }
      },
      {
        "name": "other-mcp-server",
        "transport": {
          "type": "stdio",
          "command": "other-mcp-command"
        }
      }
    ]
  }
}

9. Success Validation Checklist

• Continue.dev config updated with correct paths
• MCP server starts manually: python -m mcp_server.main
• CLI tools work: openproblems-mcp info
• Documentation cached: openproblems-mcp download-docs
• VS Code restarted after config change
• Continue.dev sidebar shows MCP tools available
• Agent can execute spatial transcriptomics tasks
• Environment validation works
• Pipeline creation and validation functional

🎉 Your OpenProblems MCP Server is now integrated with Continue.dev for powerful spatial transcriptomics AI assistance!