The Apps In ChatGPT Playbook
Apps SDK, MCP & ACP Strategy for Brands (2026)
Innovation
- By Fantasy
- 10.06.2025
OpenAI launched apps inside ChatGPT in October 2025, and the platform has evolved rapidly since. How can your brand prepare for integration with the Apps SDK, MCP, and agentic commerce?
In July 2008, Apple opened the App Store to third-party developers. Within months, every business was asking the same question: "What should our iPhone app be?" The platform shift happened faster than anyone predicted, creating entirely new ways for companies to reach customers.
We're witnessing another platform shift of equal magnitude. OpenAI's launch of apps in ChatGPT signals the start of websites playing a supporting role as AI-native interfaces become an increasingly common first stop for customer interaction. ChatGPT is becoming a major front door of the internet, and the companies that understand this shift first can capture outsize advantage in what we're calling the agentic era.
Fantasy has been preparing our partners for this transition.
The question is how quickly your brand can move before competitors build durable distribution in a landscape where customer relationships increasingly happen through AI-mediated conversations rather than traditional web browsing.
With Apps inside ChatGPT, brands will design experiences that appear natively, the moment a customer needs them. This fundamentally changes what it means to meet people where they are.Peter Smart – CXO, Fantasy
From components to complete experiences.
Apps in ChatGPT is OpenAI's Apps SDK, an interactive application framework that lets brands build custom experiences — branded mini-applications — inside ChatGPT conversations. Built on Model Context Protocol (MCP), the SDK provides full control over back-end logic and front-end UI, with apps rendered in three display modes: inline, picture-in-picture (PiP), and fullscreen.
As of late 2025, OpenAI uses "apps" as an umbrella term that includes both interactive Apps SDK apps and connected apps (formerly called "connectors") that bring external tools and data into ChatGPT for search, deep research, sync, and actions. Availability varies by app, plan, and region, so brands should define which app surface they mean before planning distribution.
OpenAI's launch partners illustrated the concept early. When someone asks ChatGPT about learning machine learning, Coursera's interactive educational module appears natively in the conversation. When users need creative assets, Canva materializes to generate presentations and marketing materials. When researching real estate, Zillow's interactive maps and property listings surface seamlessly.
These individual apps are the beginning. The larger transformation happens when multiple brand experiences chain together inside a single conversation. ChatGPT becomes a kind of orchestrator — it can combine multiple apps and tools to form end-to-end customer journeys that replace traditional website navigation entirely.
Consider someone starting a business. The conversation might begin with brainstorming, then invoke Canva to produce brand assets, followed by an app that helps choose a location, then connect to financial planning tools — all within one continuous dialogue. Each brand contributes specialized capability while ChatGPT maintains context across steps.
How the platform evolved: Plugins to Apps SDK
If your mental model still starts with "plugins," it's out of date. Here's the timeline that matters:
OpenAI wound down the plugins beta in 2024. The plugin store closed and new plugin chat creation was disabled on March 19, 2024, with existing plugin conversations continuing until April 9, 2024. OpenAI directed builders to create Actions inside a GPT instead.
In early 2024, OpenAI expanded distribution through the GPT Store, reporting over 3 million custom GPTs created within roughly two months of the GPTs feature being announced.
By October 2025, OpenAI launched apps in ChatGPT with the Apps SDK, positioning the audience as over 800 million ChatGPT users and emphasizing discovery-by-suggestion and invocation-by-name.
In December 2025, OpenAI renamed "connectors" to apps, unifying interactive apps and connected apps under one umbrella.
By early 2026, OpenAI implemented the MCP Apps UI standard for portability, and approved apps began converting to Codex plugins for distribution within Codex.
What the Apps SDK means technically in 2026
Apps built with the Apps SDK use MCP (Model Context Protocol) to connect to ChatGPT. Building an app requires an MCP server that defines tools, plus an optional web UI rendered in a sandboxed iframe when a richer interface is needed.
OpenAI's platform direction emphasizes portability. ChatGPT implements the MCP Apps UI standard, with UI communicating through a standardized JSON-RPC bridge. OpenAI-specific extensions remain available, but OpenAI advises using standard methods by default so apps can work across MCP-compatible hosts.
OpenAI's documentation explicitly states Apps SDK support is "here to stay" with "no plans to deprecate," while encouraging builders to lead with MCP Apps standards for cross-host portability.
For distribution, OpenAI's Apps SDK documentation describes a dashboard-based review flow for public distribution, with apps published into the ChatGPT Apps Directory. As of March 2026, approved apps are also converted to plugins for Codex distribution.
Commerce reality check: ACP, Checkout, and what most brands should expect
Commerce inside ChatGPT is emerging through the Agentic Commerce Protocol (ACP), an open standard developed by OpenAI and Stripe (currently in beta). OpenAI's messaging has increasingly emphasized product discovery as the near-term value — richer, visually immersive shopping with product comparisons and merchant integration.
For checkout, OpenAI's monetization guidance recommends external checkout as the generally available approach. Instant Checkout exists for select approved partners in a limited beta, currently focused on physical goods. Most brands should plan around driving strong discovery and comparison experiences inside ChatGPT, with purchase completion happening through their existing checkout infrastructure.
The CDO, CPO, and CMO’s Guide to Apps inside ChatGPT
Frequently asked questions
Apps inside ChatGPT are interactive applications built with OpenAI's Apps SDK that run natively within ChatGPT conversations. Built on Model Context Protocol (MCP), the SDK provides full control over back-end logic and front-end UI, letting brands deliver their services and products directly within the AI interface where customers are already engaged.
For brands, the practical meaning is that parts of your product can become in-conversation capability: the model can recommend your app when intent is detected, or users can call it by name. Apps support inline display, picture-in-picture, and fullscreen layouts, with context passed back to ChatGPT so the AI always understands what users are interacting with.
As of late 2025, "apps" also includes connected apps (formerly "connectors") that let ChatGPT securely reference and act on external tools and data — so teams should define whether they mean interactive Apps SDK apps, connected apps, or both when planning their strategy.
Apps inside ChatGPT transforms your brand from a destination into an in-conversation capability that meets customers where they already are. Your website becomes the structured data repository feeding AI-native experiences, while customers interact with your brand through conversational components at the moment of intent.
This shifts competition toward utility within a workflow. OpenAI's UX guidance stresses that successful apps do at least one job better because they live in ChatGPT, and explicitly warns against patterns that feel like ports of websites or ad space. The apps that succeed deliver focused value — a product finder, a calculator, a booking tool — at the precise moment the customer needs it.
The competitive advantage compounds over time. Brands that master AI-native experiences deliver personalization and contextual relevance that static websites cannot match. Every conversation becomes an opportunity to learn, adapt, and provide increasingly relevant value.
The Apps SDK has been in preview since October 6, 2025, with access expanding to Business, Enterprise, and Edu plans. Brands building during this preview period gain real expertise advantages — they understand conversation design, have optimized their data for AI consumption, and have built customer relationships through AI interfaces while competitors are still scoping.
Two realities are already clear from OpenAI's guidance. First, discovery depends heavily on the tool and metadata surface you provide — this requires iteration, not a single engineering sprint. Second, reliability depends on repeated testing (direct prompts, indirect prompts, and negative prompts) rather than one-time QA.
OpenAI's Developer Mode is the standard way to test MCP apps end-to-end inside ChatGPT. It enables read and write tool access and carries elevated risk, which means testing needs proper safeguards from day one.
Four high-value patterns have emerged across early Apps SDK implementations:
Product discovery and comparison — Shopping and discovery flows benefit from structured data and UI that helps users compare and refine options. OpenAI's March 2026 shopping updates highlight product discovery and comparisons powered by ACP. Target and DoorDash have both announced app experiences inside ChatGPT for conversational shopping and grocery ordering.
Creative production — Canva inside ChatGPT focuses on going from idea to finished creative output (presentations, social posts, marketing materials) in one conversational flow. Adobe has positioned Photoshop, Express, and Acrobat usage inside ChatGPT as well.
Learning and guided support — Coursera positioned its ChatGPT app as a way to bring trusted educational content into the conversation, making learning more accessible and contextual.
Customer operations and CX — Connected apps bring support and operational data into ChatGPT. Intercom, for instance, provides secure access to conversations, tickets, and user data inside ChatGPT for Business, Enterprise, and Edu plans.
On reach: OpenAI positions Apps SDK builders as reaching over 800 million ChatGPT users, though availability for specific apps varies by plan and region.
Preparation requires strategy matched with technical readiness. OpenAI's developer guidance has matured into concrete playbooks that change how product and marketing teams should scope and ship.
Start from real intent. OpenAI's planning docs recommend writing direct prompts (users name your app), indirect prompts (users describe the goal without naming you), and negative prompts (scenarios where your app should not activate) as an evaluation set before you build. This "golden prompt set" becomes your ongoing regression test.
Treat metadata like product copy. ChatGPT decides when to surface your app based on tool names, descriptions, and parameter documentation. OpenAI explicitly advises iterating on these because they are the primary discovery lever. Use formats like "Use this when..." plus disallowed cases to reduce accidental activations.
Design your tool surface with risk annotations. OpenAI's guidance calls out annotation hints like
readOnlyHint,destructiveHint, andopenWorldHintto help clients apply appropriate confirmations and reduce risk.Operationalize testing in Developer Mode. Validate discovery and behavior by running your golden prompt set through Developer Mode and the API Playground. Record which tool was selected, what arguments were passed, and whether the component rendered correctly.
Traditional digital strategy optimizes for attention capture and funnel conversion — page views, bounce rates, and navigation paths. Apps inside ChatGPT introduces a fundamentally different interaction model: conversational context, model-driven tool selection, and UI components deployed at the moment of need.
The unit of design changes. OpenAI's UX principles emphasize time-bound, conversational tasks with clear CTAs, while explicitly warning against re-creating long, multi-step workflows or using the UI space for ads or irrelevant messaging.
Budget allocation shifts accordingly. Rather than paying to interrupt customers with ads, you invest in being the app ChatGPT selects when customers express relevant needs. Success comes from being genuinely useful within existing workflows — a shift from brand perception as the differentiator to brand utility.
Start with a narrow job-to-be-done. OpenAI's guidance focuses on translating your product's strengths into well-scoped capabilities that improve conversations — focused utility, not feature sprawl.
Design around display modes. OpenAI supports inline, picture-in-picture, and fullscreen. PiP is useful for ongoing or live sessions that run alongside conversation; fullscreen is for deeper engagement. Choose the mode that fits the task, and design the UI to feel native to the conversational context.
Design for discovery and correctness. Treat tool descriptions and parameter docs as part of UX. Use clear, specific language: describe what triggers the tool, what it does, and what it should not be used for. This reduces accidental activations and builds user trust.
Build for portability. OpenAI's current docs recommend using MCP Apps standard keys wherever possible, and layering ChatGPT-specific extensions only when needed. This protects your investment as the MCP ecosystem expands beyond ChatGPT.
Plan for progressive engagement. Design apps that handle simple queries gracefully while supporting complex interactions. A customer might start with a basic product question and progress to configuration and purchase within the same flow.
AI-native experiences require new success frameworks. OpenAI's documentation emphasizes iterative testing and regression readiness, which implies success measurement needs both product metrics and model-behavior metrics.
Discovery quality — How often your app activates on intended prompts (recall) and stays inactive on irrelevant prompts (precision). Measure via golden prompt tests run regularly as the model evolves.
Task completion quality — Whether the user reaches the intended outcome with minimal backtracking, both in the in-chat UI and any handoff steps. OpenAI's UX principles treat time-bound, visually summarizable tasks as the ideal pattern.
Trust and safety performance — Confirmation accuracy for write and destructive actions, alignment with least-privilege design. Developer Mode carries explicit warnings about elevated risk on write actions, and the Apps SDK security guidance treats all apps as production software requiring proper security review.
Relationship progression — How customer interactions evolve over time, from simple queries to complex problem-solving, indicating growing trust and utility.
The industries that will transform first
Industries where customers already talk their way into decisions — planning, comparison, configuration, support — are positioned for early transformation.
Travel and hospitality — OpenAI launched with travel partners including Booking.com and Expedia in the initial cohort, highlighting comparison and planning flows inside chat. The entire journey from inspiration through booking can happen within ChatGPT.
Education and training — Coursera positioned its ChatGPT app as a first-of-its-kind embedded learning surface, designed to make trusted educational content accessible directly in conversation. Professional training, skill assessments, and certification programs can all embed naturally into career development discussions.
Creative tooling — Canva's ChatGPT app plus MCP server positioning signals a "design inside the conversation" workflow shift. Adobe has described Photoshop, Express, and Acrobat usage inside ChatGPT as well, covering the creative production spectrum.
Retail and e-commerce — Target and DoorDash have both announced app experiences inside ChatGPT for conversational shopping and grocery ordering. OpenAI's March 2026 commerce updates emphasize product discovery and comparisons powered by ACP. The browsing model is giving way to conversational product finding.
Customer support and CX — Intercom provides secure access to conversations, tickets, and user data inside ChatGPT for certain business plans. As more CX platforms build connected apps, support workflows increasingly happen where customers already are.
Financial services and healthcare — These categories represent natural fits for conversational interfaces — financial planning, loan comparison, symptom assessment, appointment scheduling — though they require extra care around compliance, accuracy, and trust-building within the AI context.
Beyond Apps in ChatGPT: The orchestrated web
Apps in ChatGPT represents an early step toward a more orchestrated web — where conversations can pull structured capabilities from multiple services in sequence. OpenAI's platform guidance increasingly centers standardization through MCP and MCP Apps so these experiences can travel across compatible hosts.
This evolution will likely move faster than the mobile shift. Distribution happens within an already-massive consumer interface, and the infrastructure is standardizing around open protocols rather than proprietary SDKs. The brands that invest in MCP-based architecture now build portability into their AI distribution strategy from day one.
The implications extend beyond customer-facing experiences. Internal business systems will increasingly integrate with AI interfaces, creating seamless workflows between customer conversations and business operations. Marketing, sales, customer service, and product development converge within AI-native platforms.
For marketing and product teams, this means developing new competencies. Tomorrow's brand builders will think like product managers — designing experiences rather than campaigns, optimizing for utility rather than attention, measuring discovery quality rather than impression volume.
Getting ready for the Apps in ChatGPT transition
The brands that thrive in the AI-native era will share several characteristics:
- 01.They scope for utility that fits conversation. OpenAI's UX principles explicitly reward apps that do at least one job better because they live in ChatGPT.
- 02.They invest in metadata, evaluation prompts, and iteration; treating discovery engineering as a continuous discipline rather than a launch checklist.
- 03.They treat safety, privacy, and permissions as core product features, not afterthoughts. The Apps SDK security guidance and Developer Mode warnings are explicit about elevated risk and the production reality of prompt injection.
Where should I start with Apps in ChatGPT?
Start by asking fundamental questions about your customer relationships:
- 01.What problems do your customers really need solved?
- 02.How could AI-native experiences deliver solutions more efficiently than your current web-based methods?
- 03.What unique data or capabilities could your brand contribute to customers' AI-assisted workflows?
Operationally, begin with a testable intent set and a tool surface plan. Draft direct, indirect, and negative prompts. Sanity-check tool coverage against real customer scenarios. Then test in Developer Mode using golden prompts and regression checklists.
Treat Apps in ChatGPT as a new interaction model with new constraints. OpenAI warns against replicating long-form websites or using the space for ads, and pushes builders toward small, composable actions and clear in-chat UX. The brands that internalize this design philosophy early will have a durable advantage as the ecosystem scales.
Our partners are already preparing their strategies for Apps in ChatGPT.
If you’re ready to build yours, let’s talk.