I think Retool might benefit a lot by having the possibility to expose a remote MCP server to provide data to external LLM clients.
My use case is that I have customers who use Retool as a central hub where data comes from different sources, apps to manage/observe data and information, workflows to do ETL processes, AI in the mix, etc.
But they are starting to want to use MCP-compatible LLM clients (Claude/ChatGPT) to ask something on THEIR data.
So instead of forcing them to use a chatbot within a Retool app, they might use other chatbots that, in general, are more powerful.
I’ve been thinking about this as well, and even created a remote server using Cloudflare's mcp’s, adding some tools that connect to either retools’ db or other postgres instance.
What kind of data and with what kind of structure do you think retool could expose to a mcp server? Asking just out of curiosity as I know you’ve thought this through.
Great that you did something related. I also thought to create an MCP wrapper for Retool Workflows, but I think a native implementation would be definitely better.
Yes, I know that an easy way would be to connect the db via MCP, but that way would make the user asking very detailed (and to some extent technical) questions in order to generate the right SQL.
Also, the possibility that the model doesn’t fully understand the relation between the question and the underlying data structure exists.
For example, I might have a projects table that requires several data joins, or decorations from other sources, and custom logic to include/exclude records.
That complexity might be nicely solved in a workflow that might be called by an LLM client.
So, from a question: “Show me the latest project updates” the result might be very different from an accuracy perspective if it’s an SQL query or a Workflow call.
So, to me it would open very interesting opportunity of cross integration.
I was going to say great idea and I can make a feature request for this, but it looks like we already have a ticket for this.
It is definitely on our AI roadmap, I can add +1s for you both and link this thread to that ticket so that I can share any updates we have for this functionality
