MCP servers and clients
Model Context Protocol (MCP) is an open standard that defines how AI applications discover and call tools. OpenSearch supports MCP using the following components.
| Component | Use case | Where it runs |
|---|---|---|
| MCP connector | Let OpenSearch agents call tools hosted on external MCP servers. OpenSearch acts as an MCP client. | In your cluster |
| In-cluster MCP server | Let external MCP clients call tools registered in your cluster. OpenSearch acts as an MCP server. | In your cluster |
| Standalone MCP server | Let external AI assistants query OpenSearch using its REST APIs. | Outside your cluster |
The standalone MCP server is the open-source OpenSearch MCP Server project.
Choosing an MCP server
To let external AI applications access OpenSearch, choose an MCP server based on your requirements.
| Requirement | In-cluster MCP server | Standalone MCP server |
|---|---|---|
| Setup | Enable the plugins.ml_commons.mcp_server_enabled setting. No installation is required. | Install the server and run it as a separate process. |
| Exposed tools | Tools that you register in the cluster | Tools provided by the server, which call OpenSearch REST APIs |
| Transport | Streamable HTTP | Standard input/output (stdio), Streamable HTTP, and legacy Server-Sent Events (SSE) |
| Clusters | The cluster that runs the server | One or more clusters, including Amazon OpenSearch Service domains and Amazon OpenSearch Serverless collections |