Description
Problem Statement
With the recent surge in Large Language Models (LLMs) and AI agents, there's an increasing demand for AI systems to access the latest data from various sources, including databases and public APIs. The Model Context Protocol (MCP) has emerged as a universal, open standard for facilitating these connections. Skills is a centralized repo to accommodate various tools and it's a good opportunity to support MCP client in the repo to give user the capability to connect to their own MCP servers to perform different automated tasks. The proposal to convert OpenSearch APIs to tools are covered in this RFC: opensearch-project/OpenSearch#17577
Proposed Solutions
- Configure an agent that to perform a specific/general task:
POST /_plugins/_ml/agents/_register
{
"name": "my MFC demo agent",
"type": "flow",
"description": "this is a simple demo agent that uses MFC client",
"llm": {
"model_id": "{your_model_id}",
"parameters": {
"max_iteration": 5,
"stop_when_no_tool_found": true,
"response_filter": "$.completion"
}
},
"tools": [
{
"type": "MFCTool",
"name": "MFCTool",
"parameters": {
"method": "index/_mapping",
"request_body": ${parameters.params}
}
}
],
"app_type": "my_app"
}
- Execute the agent to solve a complex task:
POST /_plugins/_ml/agents/{agent_id}
{
"parameters": {
"question": "Please help me create a new index in my cluster",
"params": "{ \"mappings\": { \"properties\": { \"text\": { \"type\": \"text\" } } }, \"settings\": { \"index\": { \"number_of_replicas\": \"1\", \"number_of_shards\": \"2\" } }}"
}
}
- During the agent running, the MFCTool get invoked and the
paramsin theparametersget passed to corresponding MFC server and the task complete.
This tool is generic enough to support different MFC server interaction with different request body, and all the MFC protocol details will be handled by OpenSearch.