ChatGPT Work Connectors: The 1,400-App Directory (2026)
Connector breadth was the headline of the ChatGPT Work launch on July 9, 2026, and it’s the strongest thing about the product. Work can @-mention apps from a directory of 1,400+ connectable apps to pull context into a run, and OpenAI migrated that directory to the Plugins surface the same day. On Business plans there are 60+ data connectors with admin governance around them.
But there’s a distinction inside those numbers that decides whether a connector is actually useful to you, and it’s not the one people check.
Reading is not acting
Most of the 1,400 are context sources. Work can @-mention them to read what’s there (pull the deal record, read the ticket, look at the doc), and that genuinely improves output quality, because the agent isn’t guessing at your business.
Writing back is a narrower set with more friction. Work prompts for approval on write actions, which is the correct default for an autonomous agent and also the thing that quietly breaks unattended use. A run that pauses for your approval isn’t running while you’re asleep; it’s waiting.
So when you’re evaluating whether a connector solves your problem, the real question isn’t “is my app in the directory.” It’s: do I need the agent to read this, or to change it? Directory presence answers the first. It doesn’t answer the second.
What’s covered well
The mainstream business stack is solid: the tools most companies actually run on. Launch testers wired Work into CRM, call-recording, messaging, and project-management tools, which is a fair picture of where the depth is. Business tiers add the admin layer: which connectors are enabled, for whom, with SOC 2 Type 2 and SSO around them.
The enterprise data platforms have dedicated coverage too, which is why we have individual write-ups on the Salesforce, SharePoint, Snowflake, Tableau, ServiceNow, Workday, SAP, Power BI, Box, and Google Workspace connectors.
Where the directory runs out
Vertical and long-tail software. A directory that size still misses the specialized tools smaller businesses run on: field service, legal practice management, property management, clinical scheduling. If it’s not listed, the fallback is building an MCP server yourself.
Anything needing a standing connection. Connectors feed a run that’s already started. None of them make Work start. There’s no connector that means “when this changes, go”. That’s a trigger, and Work doesn’t have them.
Write depth varies. Being connected tells you nothing about whether the agent can create, update, or delete in that system, or how much approval friction sits in the way.
How this compares to Carly
Different model, worth understanding rather than just counting.
Carly has 260+ native integrations across 45+ categories, a smaller directory than 1,400 and deliberately a different kind of thing. These are toggle-on connections built to act: send the email, book the meeting, update the record, file the document, with no per-action approval queue. Anything outside the native list connects with your own API key, pasted at carlyassistant.com/integrations, so an app being absent from the catalog isn’t a dead end.
The other difference is direction. Carly’s integrations are triggers as well as destinations: a new row, a changed deal stage, an inbound email can start work, not just inform it. That’s the capability the 1,400-app directory doesn’t contain at any size.
And you can point ChatGPT at Carly. Carly runs an MCP server at https://carlyassistant.com/mcp/, which you add under Plugins in ChatGPT. That gives a Work run direct access to Carly’s email, calendar, contacts and CRM, workflow, file, notes, and booking tools, so the run can finish an action through Carly instead of stopping at a draft with an approval prompt. Claude connects the same way under Customize → connectors.
FAQ
How many connectors does ChatGPT Work have?
The app directory covers 1,400+ connectable apps for pulling context, migrated to the Plugins surface at the July 9, 2026 launch. Business plans include 60+ governed data connectors with admin controls.
Can ChatGPT Work write to my apps, not just read them?
For some connectors, yes, but write actions prompt for approval, so unattended runs stop and wait. Directory listing indicates the app can be read from; it doesn’t guarantee write depth.
What if my app isn’t in the ChatGPT Work directory?
The supported path is building or hosting an MCP server and connecting it as a custom connector. That’s real engineering work, not a toggle.
Can a connector trigger a ChatGPT Work run?
No. Connectors supply context to a run you’ve already started, or one a schedule started. There are no webhooks or event triggers.
More: What is ChatGPT Work · ChatGPT Work pricing · ChatGPT connectors · ChatGPT Work alternatives · Carly integrations · Best AI workflow automation tools
Ready to automate your busywork?
Carly schedules, researches, and briefs you—so you can focus on what matters.
See what people say
"Before Carly, I relied on a Calendly link, but the whole process felt impersonal and not very professional. Carly changed that by handling all the back-and-forth, so I'm no longer stuck in endless email threads trying to line up schedules.
Now Carly reaches out to candidates, shares my real-time availability, lets them pick a slot, then sends a Zoom link and drops it straight into my calendar. She sends reminders to both of us before each call, which has significantly reduced no-shows and last-minute confusion.
On top of scheduling, Carly acts like a full executive assistant, sending me my schedule the night before so I can prepare for each call. It reminds me of the old x.ai assistant, but Carly is noticeably smarter, faster, and better suited to my healthcare recruitment business."
