Hey builders! I'm a product designer who spent the last five years at a start-up but hasn’t built a single thing on my own. I'm embarrassed and ashamed, so I told myself that's going to change. So for the rest of the year and beyond, I'm focused on building. Looking forward to connecting with all of you!

I should caveat this by saying I hope this is just a simple bug that OpenAI is currently working on and has a solution quite soon, but in the meantime, this was my experience with trying to get my MCP to connect to Zapier: After wrestling with this for a few days, I finally figured out why my Zapier MCP kept failing to reconnect and throwing random CEL parse or 401 auth errors — even after deleting and rebuilding everything from scratch. Now Im not 100% certain as to why the API key doesnt always seem to work, but if you continually get CEL Parse Error or 401 Auth Errors, what worked for me was to set your credentials as a header in the MCP configuration and apply the following: Key: Authorization Value: Bearer <your_api_key> Once I saved that and re-tested the connection, it passed instantly. No more CEL errors, no 401s — clean handshake and data started flowing again. If you’re seeing the same problem (connection test failing after a token refresh), try switching to header-based auth. It looks like Agent Builder currently parses the Access Token field incorrectly in some cases, especially after key rotation. Next is to figure out how to keep this connection going cause from my understanding the Auth Tokens need to be rotated every 24-48 hours which is turbo annoying. Anyways, hope that helps someone.

Has anyone else run into problems where the Zapier MCP server connection just stops working and refuses to reconnect? Everything was running perfectly in my workflow until the authentication token expired. Normally I’d just reconnect, but this time it completely broke. When I try to reconnect or rebuild the MCP: - The connection test fails immediately — sometimes with a CEL parse error if I rotate or re-enter the API key. - Other times I get a 401 authorization error, even after deleting and recreating both the MCP server in Agent Builder and the integration in Zapier. - The connection used to work fine — same workspace, same setup. At this point I’m wondering whether: - Zapier changed something on the MCP side, - the OpenAI Agent Builder client/runtime changed how it handles Zapier MCP auth. I did notice that Zapier now has two clients, OpenAI Agent Builder and OpenAI API. Im using the OpenAI Agent Builder client which I assume is correct. - there’s some hidden credential cache I need to clear. Has anyone else seen this or found a reliable way to get the Zapier MCP to reconnect after a token expires?

I’m trying to confirm something about how Agent Builder handles outbound HTTP requests. If I want to make a simple GET request from within the Agent Builder, do I have to route it through “Webhooks by Zapier: GET”, or is there a more direct / native node for making HTTPS requests? Example use case: I’d like the agent to query the DNS-over-HTTPS API at dns.google (or some other DNS lookup tool) Should I create a Webhooks by Zapier GET MCP for that, or is there some workaround to make this call directly inside the agent with some built-in “Web Request”? I dont think thats possible but maybe theres some hack that I havent discovered yet. Thoughts?

Hi all, has anyone actually even managed to deploy or host your agent builder workflows? i see so many people posting about it, but no one has an actual deployment / instance that is accessible.. is it even supported?