Web Services ChatGPT Plugins
Last reviewed
May 9, 2026
Sources
No citations yet
Review status
Needs citations
Revision
v2 ยท 2,499 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Last reviewed
May 9, 2026
Sources
No citations yet
Review status
Needs citations
Revision
v2 ยท 2,499 words
Add missing citations, update stale details, or suggest a clearer explanation.
See also: ChatGPT Plugins, ChatGPT Plugin Categories, Web Development ChatGPT Plugins, Search Engine ChatGPT Plugins and Internet ChatGPT Plugins
Web services ChatGPT plugins were a category of third party extensions for ChatGPT that connected the chatbot to remote application programming interfaces, software as a service products, business automation suites, and cloud platforms. They formed one of the broadest groupings inside the original ChatGPT Plugins catalogue during its short lifespan from March 2023 to April 2024, sitting at the boundary between productivity tooling, customer relationship management, payments, communication, and developer infrastructure. Unlike the Search Engine ChatGPT Plugins grouping, which mainly fetched and summarised public web pages, the web services category exposed authenticated user accounts, multi step workflows, and write actions inside live business systems.
The "web services" tag was a fuzzy bucket. It covered tools that called hosted APIs to perform real work in third party software, including spreadsheet automation, customer messaging, scheduling, payment processing, document storage, and cross application orchestration. Many entries also appeared under adjacent tags such as productivity or developer tools, so the same plugin could be discovered through several paths. The category is distinct from Web Development ChatGPT Plugins, which focused on building, testing, and scraping the public web rather than driving SaaS dashboards.
OpenAI announced ChatGPT plugins on March 23, 2023, describing them as tools that would help ChatGPT "access up-to-date information, run computations, or use third-party services."[1] At launch the company shipped two of its own first party plugins, a code interpreter and a web browsing tool, alongside an inaugural cohort of third party plugins from Expedia, FiscalNote, Instacart, KAYAK, Klarna, Milo, OpenTable, Shopify, Slack, Speak, Wolfram, and Zapier.[1][2] Several entries in this first cohort sat squarely inside the web services category because they wrapped large hosted business platforms rather than search engines or static knowledge bases.
Access to plugins began in alpha for a limited group of developers and ChatGPT Plus subscribers on a waitlist. On May 12, 2023, OpenAI began rolling the plugin store out more broadly to ChatGPT Plus subscribers and removed the waitlist over the following week.[3] By that point the catalogue contained more than 70 third party plugins. The store kept growing through the second half of 2023, eventually exceeding 1,000 plugins before OpenAI announced its end. On March 19, 2024, OpenAI disabled the creation of new plugin conversations and stopped accepting new plugin installations. Existing plugin conversations continued working until April 9, 2024, when the entire plugin platform was shut down.[4][5]
During the same period, OpenAI introduced Custom GPTs on November 6, 2023 at the DevDay developer conference, and the GPT Store opened to the public on January 10, 2024. Custom GPTs replaced plugins with a similar mechanism, called Actions, that calls remote APIs through OpenAPI specifications. Most viable web services plugins migrated to either a Custom GPT, a built in feature inside their own product, or a workflow inside an external automation tool.
The web services tag covered five overlapping use cases. The boundary between them was never strict, and several plugins straddled two or three. The table below summarises the typical shape of each subgroup.
| Use case | What the plugin did | Representative plugins |
|---|---|---|
| Cross app automation | Triggered prebuilt workflows that connected dozens or thousands of SaaS products | Zapier |
| Team communication | Read, summarised, and posted messages inside a workspace | Slack |
| Customer and commerce | Looked up orders, recommended products, processed payments, or fetched store data | Klarna, Shopify, Instacart |
| Reservations and bookings | Searched availability and held bookings inside hosted platforms | OpenTable, KAYAK, Expedia |
| Cloud and developer infrastructure | Surfaced data from hosting, networking, or developer platforms | Cloudflare Radar, Cloudflare Docs |
The shared trait across these subgroups was that the plugin sent the chat session into a live remote system instead of a static reference data source. That distinction is what separated the web services category from search and computational categories, where Wolfram or a browse plugin merely returned text the model could read.
Every plugin in the store followed the same architecture. The developer hosted an HTTPS API endpoint plus a small manifest file at the path /.well-known/ai-plugin.json on the same domain. The manifest declared the plugin name, description, authentication method, and a pointer to an OpenAPI specification that listed each available action.[6] When a Plus user installed the plugin, ChatGPT downloaded the manifest, validated the OpenAPI document, and added the plugin's action descriptions to its tool list.
During a chat session, the model decided when to call the plugin based on the natural language descriptions in the OpenAPI document. For web services plugins this often involved an authentication step, since the plugin needed to act on behalf of a specific user. The platform supported four authentication modes: no authentication, service level keys, user level OAuth, and user level keys.[6] Most production web services plugins used OAuth, which bounced the user through the partner's normal login flow during plugin installation.
A single conversation could enable up to three plugins at the same time. ChatGPT chose which plugin to call based on the user's prompt and the action descriptions, and it could chain multiple calls in sequence. The model returned the API response to the user inside the chat, often after some natural language summarisation.
Five of the twelve inaugural third party plugins announced on March 23, 2023 fit cleanly inside the web services category. They are summarised below with the public description that OpenAI and the partner used at launch.
| Plugin | Operator | Function at launch |
|---|---|---|
| Zapier | Zapier | Triggered preconfigured Zaps across more than 5,000 connected apps such as Google Sheets, Gmail, HubSpot, and Salesforce[7] |
| Slack | Salesforce / Slack | Summarised channels, drafted replies, and helped research topics inside a Slack workspace[8] |
| Klarna | Klarna | Returned product recommendations and price comparisons through Klarna's shopping search and compare service[2] |
| Shopify | Shopify | Searched millions of products across Shopify storefronts on behalf of the user[1] |
| FiscalNote | FiscalNote | Provided real time legal, political, and regulatory data through a curated set of FiscalNote APIs[9] |
The Zapier plugin became the most discussed entry in the category. Because Zapier itself was a connector layer over thousands of SaaS products, installing the single Zapier plugin gave a Plus user indirect access to roughly 5,000 downstream apps and tens of thousands of prebuilt actions.[7] The plugin let the user describe a desired action in plain English, and Zapier mapped it to a Zap configured in the user's Zapier account.
The Slack plugin emerged from a separate partnership announcement on March 7, 2023, when Salesforce and OpenAI introduced the ChatGPT app for Slack alongside Salesforce's Einstein GPT for customer relationship management.[8] The plugin allowed users to summarise channels, generate draft replies, and research topics without leaving Slack.
After the May 2023 broader rollout, the catalogue expanded quickly. By mid 2023 the store contained more than 200 plugins, and by the end of 2023 it exceeded 1,000.[3] Cloudflare announced two ChatGPT plugins on May 15, 2023: a Cloudflare Radar plugin that surfaced live internet traffic and attack data, and a Cloudflare Docs plugin built on a retrieval pattern over its developer documentation.[10] Both marked the entry of a major cloud and content delivery operator into the plugin store, and Cloudflare published the open source Workers based template that other developers used to build their own plugins.
Third party developer platforms also released plugins inside the web services tag. AskYourPDF let Plus users upload or link a PDF and chat with its contents through a hosted vector database. Bramework offered a hosted SEO research workflow over its keyword and ranking data. A long tail of smaller plugins wrapped commercial APIs for tasks like email drafting, calendar management, document storage, and form generation, and many later relaunched as standalone Custom GPTs after the deprecation.
Several integrations widely associated with ChatGPT in 2023 were not formally part of the plugin store. Salesforce's Einstein GPT and Slack GPT called OpenAI APIs from inside Salesforce products and were not installable plugins. Twilio, HubSpot, and similar providers offered ChatGPT integrations through external automation platforms such as Zapier or Pipedream rather than as direct first party plugins. The historical record is cleaner if the formal web services plugin list is limited to entries that shipped with their own manifest in the store.
Because the plugin store used loose tagging, several categories overlapped with web services. The table below shows where the boundary fell during the active period of the store.
| Adjacent category | Primary purpose | Overlap with web services |
|---|---|---|
| Search Engine | Query the open web and return cited snippets | Both could fetch live data, but search plugins were stateless and unauthenticated |
| Web Development | Build, test, scrape, or visualise web pages | Web development plugins targeted the public web, while web services plugins targeted authenticated SaaS dashboards |
| Internet | Generic browsing and link reading | Internet plugins fetched arbitrary URLs without an API contract; web services plugins called typed APIs |
| Productivity | Calendars, notes, email | Many web services plugins also appeared under productivity because the underlying SaaS product was a productivity tool |
| Shopping | Product search and recommendations | Klarna, Shopify, and Instacart appeared under both shopping and web services |
The practical effect of this overlap was that a Plus user looking for, for example, a Slack workflow might find the same plugin under web services, productivity, and communication tags. The store offered no formal taxonomy, and the categorisation was driven by the plugin developer's own metadata.
OpenAI did not publish detailed usage statistics for plugins, but several public statements indicated that adoption was concentrated among a small set of power users. Most ChatGPT Plus subscribers either never enabled plugins or installed one or two and rarely returned. Inside the web services category, Zapier was widely cited as the most installed plugin, followed by Slack and Wolfram. Plus users had to enable plugins in beta features, install each plugin separately, then remember which three to activate per chat, which created friction for casual use.
For power users, the web services category had attractive properties. Zapier's plugin let a single chat session reach into thousands of downstream tools without the user touching a dashboard, which made it especially popular for ad hoc automation. Slack's plugin was widely used for meeting summarisation and inbox triage.
Web services plugins raised distinct security questions because they often acted on behalf of an authenticated user inside a live system. OpenAI's plugin documentation required developers to publish a privacy policy, declare whether the plugin had write access, and specify data retention behaviour.[6] Plus users saw a confirmation prompt the first time a plugin attempted a sensitive action in a session.
Researchers and OpenAI's own safety team highlighted several risks. A poorly described OpenAPI document could trick the model into calling an unintended action. A malicious web page returned through a browsing plugin could attempt prompt injection that pointed at a separate web services plugin in the same chat. The platform mitigated these by limiting plugins to three per chat and requiring fresh confirmation for write actions, but the underlying class of risks contributed to the eventual move toward Custom GPTs. Prompt engineering practices for web services plugins differed from chat only practices because the user often wanted predictable structured behaviour, and users learned to phrase requests in ways that mapped cleanly onto a single plugin action.
OpenAI began signalling the end of plugins in late 2023 once Custom GPTs were released. The official sunset notice arrived in early 2024. The timeline was as follows.
| Date | Event |
|---|---|
| March 23, 2023 | Plugins announced with twelve inaugural third party partners[1] |
| May 12, 2023 | Broader rollout to ChatGPT Plus subscribers, waitlist removed[3] |
| November 6, 2023 | Custom GPTs announced at OpenAI DevDay |
| January 10, 2024 | GPT Store opens to all ChatGPT Plus subscribers |
| March 19, 2024 | Plugin store closed to new conversations and new installs[4] |
| April 9, 2024 | All plugin conversations stop working[4][5] |
Most active web services plugin developers migrated to Custom GPTs. Zapier built a successor called AI Actions that exposed the same connector library through a Custom GPT. Salesforce and Slack pivoted toward Slack GPT and the broader Salesforce AI strategy, which embedded large language model features directly inside their products. Klarna, Shopify, and Instacart kept Custom GPT entries on the GPT Store. FiscalNote reused its plugin OpenAPI definitions inside Custom GPTs, and Cloudflare invested in its open source plugin and AI gateway infrastructure, which fed into later Workers AI and AI Gateway products.
The web services category is historically important even though the platform itself was short lived. It marked the first widely deployed example of a consumer chatbot calling typed third party APIs on a user's behalf, and the pattern foreshadowed the broader Actions and Assistants framework that OpenAI and other vendors built throughout 2024. Most modern agent frameworks borrow directly from the manifest plus OpenAPI structure that the plugin store popularised.
The category also exposed the practical limits of a fully open plugin marketplace inside a chatbot product. Discovery, trust, and authentication were difficult problems at scale, and the eventual shift to Custom GPTs showed a preference for curated, builder driven distribution over a general purpose store. For a wider view of the parent platform see ChatGPT Plugins and ChatGPT Plugin Categories. Adjacent groupings are documented at Web Development ChatGPT Plugins, Search Engine ChatGPT Plugins, and Internet ChatGPT Plugins. The successor distribution channels are Custom GPTs and the GPT Store.