This topic was automatically generated from Slack. You can find the original thread here.
Hi! I’m just getting started learning about Pipedream Connect. Is there a way to programmatically get the return value of an action? For example, I want to know ahead of time what the structure of the response looks like for gitlab-list-commits (GitHub). It looks pretty clear how to get the props, but not the return. Thanks!
We don’t currently expose the output schema of actions, but it’s something we’re thinking about
Ah okay, yeah I saw that some of the components had defined some metadata of the return types (pipedream/components/hubspot/common/constants.mjs at master · PipedreamHQ/pipedream · GitHub)
Yea, HubSpot specifically is one of the more complex / dynamic integrations actually. The input schema is basically fully dynamic based on the user’s account configuration in HubSpot, which makes defining the output schema tricky as well.
Makes sense. The use case we have in mind is being able to use an LLM to stitch together action calls, but it’s a bit difficult to do without having the output schemas. Do you have any suggestions that might support this better?
And specifically it’s not having an LLM make MCP calls, but more having it create a structured series of actions ahead of time
Yea makes sense, we’ve seen most customers have seen more success by running the action and processing the output, which has some pros and cons I think
Does Pipedream have a concept of which actions may have side effects so we could know which are “safe” to run without user confirmation to inspect the output? e.g. something like fetching a Hubspot account is probably safe to run
No unfortunately not, but agreed that’d be very handy…
Mm cool cool. Thanks for the helpful responses 
ya’ll have built a really cool platform
Does that feel like a deal breaker to you?
We’re evaluating a few different solutions right now and trying to figure out what our build out would look like for each one so it’s not too clear yet where the deal breakers are. Ideally we’d be able to take some user’s request and dynamically construct a series of third party integration calls with an LLM
Yea the use case makes total sense, we have some other customers working on similar products using either our MCP server or the Connect API in general
Do ya’ll have anything in the near term roadmap that would improve our use case? e.g. output schemas
It’s on the roadmap but I can’t promise it’s a quick win, so not sure how near term it’d be, realistically
Mmm makes sense
Btw just curious, how’d you learn about Pipedream / make your way to our docs?
Was using OpenAI’s DeepResearch to find and evaluate integration providers haha
curious, are all of pipedream’s supported actions also supported by the MCP implementation?
Yes