Asset Panda MCP Client Wrapper

This is a real MCP server that wraps your public HTTP API and makes it available to Claude Desktop via the Model Context Protocol.

What This Does

Claude Desktop (MCP Protocol)
        ↓
MCP Client Wrapper (this package)
        ↓ HTTP
Public API (Lambda + API Gateway)
        ↓
Bedrock Knowledge Base

Setup for Claude Desktop (Windows)

Option 1: Local Development (Current Setup)

Your Claude Desktop config at C:\Users\reymondko\AppData\Roaming\Claude\claude_desktop_config.json:

{
  "mcpServers": {
    "assetpanda-public": {
      "command": "node",
      "args": ["\\\\wsl.localhost\\Ubuntu\\home\\reymondko\\pioneer\\packages\\mcp-client-wrapper\\dist\\index.js"],
      "env": {
        "MCP_API_URL": "https://3bjnmrfh8h.execute-api.us-east-1.amazonaws.com/prod/mcp-tools"
      }
    }
  }
}

Steps:

1. Build the wrapper: npm run build (already done)
2. Restart Claude Desktop
3. Ask Claude: "Search the knowledge base for collections"

Option 2: Publish to NPM (For Sharing)

Once published to NPM as @assetpanda/mcp-client, anyone can use it:

{
  "mcpServers": {
    "assetpanda": {
      "command": "npx",
      "args": ["-y", "@assetpanda/mcp-client"],
      "env": {
        "MCP_API_URL": "https://3bjnmrfh8h.execute-api.us-east-1.amazonaws.com/prod/mcp-tools"
      }
    }
  }
}

How It Works

1. MCP Protocol (stdio communication)

Claude Desktop talks to this wrapper using the MCP protocol over standard input/output.

2. Tool Registration

The wrapper registers one tool:

• search_knowledge_base: Search Asset Panda KB
  • Input: query (required), knowledgeBaseId (optional)
  • Returns: Search results with citations

3. HTTP API Calls

When Claude calls the tool, the wrapper:

1. Receives MCP request from Claude Desktop
2. Converts it to HTTP POST request
3. Calls your public API at https://3bjnmrfh8h.execute-api.us-east-1.amazonaws.com/prod/mcp-tools
4. Returns results back to Claude via MCP protocol

Files

• index.ts - Main MCP server implementation
• dist/index.js - Built executable
• package.json - NPM configuration
• tsup.config.ts - Build configuration

Development

# Install dependencies
npm install

# Build
npm run build

# Watch mode
npm run dev

Testing Locally

Test the built wrapper directly:

export MCP_API_URL="https://3bjnmrfh8h.execute-api.us-east-1.amazonaws.com/prod/mcp-tools"
node dist/index.js

This starts the MCP server. It communicates via stdio, so you'll need an MCP client (like Claude Desktop) to interact with it.

Publishing to NPM

# Update version in package.json
npm version patch  # or minor, or major

# Build
npm run build

# Publish (requires NPM account and login)
npm publish --access public

After publishing, anyone can use:

npx @assetpanda/mcp-client

Environment Variables

• MCP_API_URL: Your public API endpoint (default: http://localhost:3000/mcp-tools)

Usage Examples

In Claude Desktop

Just ask natural language questions:

• "Search the knowledge base for collections"
• "What are dynamic collections in Asset Panda?"
• "How do forms work?"

Claude will automatically use the MCP tool.

From Your Python Agent

You don't need this wrapper - just call the HTTP API directly:

from strands_tools.http_request import http_request as http_request_tool

@tool
def search_kb(query: str):
    tool_use = ToolUse(
        id="kb-search",
        name="http_request",
        input={
            "method": "POST",
            "url": "https://3bjnmrfh8h.execute-api.us-east-1.amazonaws.com/prod/mcp-tools",
            "body": json.dumps({
                "tool": "search_knowledge_base",
                "params": {"query": query}
            }),
            "headers": {"Content-Type": "application/json"}
        }
    )
    return http_request_tool(tool_use)

Comparison: Local vs Public MCP

| Feature | Local (AWS MCP Server) | Public (This Wrapper) | |---------|------------------------|------------------------| | Setup | AWS credentials locally | Just needs API endpoint | | Authentication | AWS IAM | None (public API) | | Access | Your machine only | Anywhere with internet | | Use Case | Development | Production, sharing | | Cost | Free (local) | Lambda + API Gateway costs |

Troubleshooting

"Cannot find module" error

Make sure dependencies are installed:

npm install

"Permission denied" error

Make the file executable:

chmod +x dist/index.js

API connection errors

Check that the API_URL is correct and the API Gateway is running:

curl -X POST "https://3bjnmrfh8h.execute-api.us-east-1.amazonaws.com/prod/mcp-tools" \
  -H "Content-Type: application/json" \
  -d '{"tool":"search_knowledge_base","params":{"query":"test"}}'

Claude Desktop doesn't see the tool

1. Check Claude Desktop logs (Help → View Logs)
2. Verify the config file path is correct
3. Make sure you restarted Claude Desktop after changing config
4. Check that Node.js is accessible from Windows (run node --version in PowerShell)

API Endpoint

Your current API endpoint is:

https://3bjnmrfh8h.execute-api.us-east-1.amazonaws.com/prod/mcp-tools

This is deployed in the dev environment and searches the dev Knowledge Base (6XXUK2ADAJ) by default.