Support Multiple Prometheus Data Sources in a Single MCP Server Instance

[kubehan]

We would like to request support for multiple Prometheus data sources within a single instance of the Prometheus MCP Server.

In our environment, we have hundreds of Prometheus instances (e.g., one per cluster or tenant), but we prefer to deploy only one MCP server for operational simplicity and resource efficiency. Currently, the MCP server appears to be configured for a single fixed Prometheus endpoint.

It would be extremely valuable if the MCP server could:
Accept a list of named Prometheus data sources (e.g., via config file or environment variables),
Allow clients to specify which data source to query via an identifier (e.g., through a header, URL parameter, or context field),
Route queries dynamically to the appropriate Prometheus backend based on that identifier.

This would enable scalable, multi-tenant usage without requiring hundreds of MCP server deployments.

Thank you for considering this enhancement!

[tjhop]

Thanks for taking the time to write this up! It's helpful to hear/see how others are making use of it so I can better guide development efforts. 🙃

In our environment, we have hundreds of Prometheus instances (e.g., one per cluster or tenant), but we prefer to deploy only one MCP server for operational simplicity and resource efficiency. Currently, the MCP server appears to be configured for a single fixed Prometheus endpoint.

That's pretty reasonable/normal. And yes, the MCP server is currently designed to connect to a single Prometheus API compatible endpoint. The "API compatible" part there is the important bit -- do you have a centralized query aggregation/long term storage service you could point the MCP server to instead? Mimir, Cortex, Thanos, etc all work. I'm also working on adding support for external backends that provide custom tools specific to that service (with Thanos already supported):

https://github.com/tjhop/prometheus-mcp-server?tab=readme-ov-file#prometheus-compatible-backends

This would ideally help solve the multi-tenant usage, as well as limiting the number of MCP servers to deploy.

[yeya24]

the MCP server is currently designed to connect to a single Prometheus API compatible endpoint

I think it is worth revisiting this again. It is quite common to manage a lot of Prometheus servers the same time without a centralized aggregation service like Thanos or Cortex. If the MCP server is able to support multiple data sources it would make things much easier.

Opensearch MCP server might be a good example. It is able to support multiple connection modes. Single data source, multiple data sources and dynamic connection. https://github.com/opensearch-project/opensearch-mcp-server-py/blob/main/USER_GUIDE.md#server-modes