We started seeing the value of what MCP’s could provide, so our company produced this glue layer where we saw a lot integration difficulties. (Not to mention reinventing the wheel 1,000 times with the mcp sdk).
I am hoping that this provides value to people out there who are having to integrate many MCPs, or add MCP layers to existing systems.
One bonus thing here that isn’t obvious, is that you can use the aggregator to map between different MCP protocols. Since Cursor uses CLI and SSE, but doesn’t support http, this is a useful tool for configuring cursor to work with ANY MCP. We needed the ability to host serverless mcp’s over http in the cloud. Well, you can run an aggregator that listens on sse, and maps to one or more that are in the cloud on http, as well as running mcp’s locally.
With just a change of a config, this could be executed from cli mode.
Also another use we are using it for, is creating a bundle of mcp’s that are in a docker container, and then having a config for the mcp-aggregator (a tool auto installed from a global npm install), from any dev location, you can spin up a whole stack of MCP tools. Just include the a docker compose file and an aggregator config file in your repo, and there it is.
Something I’m looking at doing here soon, is configuring what tool’s are exposed (black/white list), to limit what any AI has access to, so it can be focused.
But after days of work, I don’t have to play around with the model context protocol library for any reason ever. WOO.