SaaS product integrations
Bring an existing service's search, data, account context, and carefully scoped actions into ChatGPT.
Independent tool overview
Apps SDK at a glance
OpenAI's Apps SDK introduced a way to build interactive integrations that run inside ChatGPT. OpenAI has since folded that model into its broader Plugins platform, which combines MCP servers, optional in-chat UI, reusable skills, authentication, and a universal directory for ChatGPT and Codex.
Visit the official Apps SDK site ↗
- Current name
- OpenAI Plugins
- Original product
- Apps SDK
- Protocol
- Model Context Protocol (MCP)
- Surfaces
- ChatGPT and Codex
- Interactive UI
- Optional sandboxed web component
- SDK fee
- No separate fee listed
Overview
What Apps SDK is
The Apps SDK is the foundation OpenAI launched for developers who want their product's data, actions, and interface to appear inside ChatGPT. It uses the Model Context Protocol (MCP): a remote server exposes well-described tools, while an optional web component can render an interactive experience in the conversation.
In 2026, OpenAI moved the Apps SDK documentation into Plugins. That is an evolution rather than a shutdown: existing app-style integrations map to the MCP-and-UI portion of a plugin, while the newer package can also include skills for repeatable ChatGPT and Codex workflows. Teams should follow the current Plugin documentation and directory rules for new builds.
Use cases
Who Apps SDK is best for
The strongest fit depends on the job you need the product to complete, not the size of its feature list.
Interactive in-chat tools
Build maps, configurators, visualizations, forms, and other experiences that need more than a text response.
Internal knowledge workflows
Connect governed company systems so approved users can find information and start repeatable workflows.
Cross-client MCP services
Expose a standards-based MCP server that can be packaged for OpenAI clients without rebuilding the underlying business logic.
Developers with an existing backend
Reuse current authentication, APIs, permissions, and premium entitlements instead of recreating the product inside ChatGPT.
Capabilities
Core Apps SDK features
MCP tools
Define structured tools that let the model retrieve data or request actions from a remote service.
Interactive components
Attach optional web UI to a tool result so users can inspect and manipulate richer content in the conversation.
Skills
The current Plugin format can bundle instructions, scripts, and resources for repeatable ChatGPT and Codex workflows.
Existing accounts and entitlements
Authentication support lets a plugin respect a product's current users, permissions, and paid features.
Read and write workflows
Tools can search connected information or propose external actions, with confirmation expected for consequential changes.
Developer testing
OpenAI provides a development connection flow for checking tools, metadata, UI, authentication, and error handling before submission.
Directory publishing
Approved packages can be distributed through OpenAI's plugin directory rather than requiring every user to configure a server manually.
Conversion specifications
OpenAI publishes structured patterns for workflows such as reservations, quotes, and product checkout.
Open implementation choices
Developers retain control of their server, data layer, business logic, and component code.
Metadata-driven discovery
Tool names, descriptions, parameters, and package metadata help the model and users understand when the integration is relevant.
Process
How the Apps SDK workflow works
Step 1
Choose a narrow job
Start with a specific user outcome and identify the minimum data and actions needed to complete it.
Step 2
Design the tools
Create clear MCP tool names, descriptions, input schemas, outputs, and read-versus-write annotations.
Step 3
Add UI only where useful
Register a component for results that benefit from controls or visualization; keep simple tasks conversational.
Step 4
Implement trust controls
Add authentication, least-privilege authorization, consent, privacy handling, and confirmations for consequential actions.
Step 5
Test realistic failures
Check adversarial prompts, permission boundaries, unavailable dependencies, duplicate actions, and mobile or narrow layouts.
Step 6
Deploy and submit
Host the MCP service securely, package current metadata and assets, submit for review, and monitor behavior after release.
Cost
Apps SDK pricing and free plan
OpenAI does not list a separate Apps SDK or Plugin SDK charge. Real costs come from the ChatGPT access required by users and from the developer's own hosting, APIs, storage, authentication, and support infrastructure.
Plugin framework
No separate SDK fee listed
Use the public documentation, MCP standards, examples, and packaging format to build an integration.
- Developer tools and documentation
- MCP server support
- Optional UI and skills
ChatGPT access
Plan-dependent
Availability of plugins, custom MCP connections, sync, research, and actions can vary by ChatGPT plan, workspace, region, and admin policy.
- Consumer and workspace eligibility varies
- Admin controls may apply
- Feature access can differ by client
Developer infrastructure
Usage-based
The developer pays for the remote server and any databases, third-party APIs, model calls, monitoring, or identity services the integration uses.
- Hosting is not included
- Third-party service fees may apply
- Operational support remains the developer's responsibility
Pricing checked . Check current pricing at the source ↗
Assessment
Apps SDK strengths and limitations
Where it stands out
- Puts a product's capabilities directly in the conversation where a user is already working.
- MCP separates the integration contract from the underlying backend implementation.
- Optional UI supports use cases that are awkward or unclear in text alone.
- Existing login, permissions, and premium entitlements can remain the source of truth.
- The newer Plugin package can combine service connections with reusable workflow instructions.
- One directory creates a clearer distribution path than asking users to configure raw endpoints.
- Structured tool schemas make capabilities easier to test and govern than prompt-only integrations.
- Developers retain ownership of their data layer and business logic.
What to consider
- Apps SDK is now legacy terminology; new implementations should follow the current Plugin documentation and packaging rules.
- A useful integration still requires a production backend, secure hosting, monitoring, and ongoing support.
- ChatGPT decides when a tool is relevant, so invocation is less deterministic than a traditional app navigation flow.
- Poor tool descriptions can cause missed calls, incorrect calls, or confusing parameter collection.
- Remote tools introduce prompt-injection, data-exfiltration, and unintended-action risks that require explicit defenses.
- Write actions need least-privilege permissions, idempotency, clear previews, and confirmation before meaningful external effects.
- Interactive components run within host constraints and must follow OpenAI's UI, security, and content rules.
- Submission approval does not guarantee featuring, discovery, or sustained usage.
- Availability can differ by plan, country, model, client, workspace role, and administrator settings.
- Authentication and account linking add abandonment points and a larger security surface.
- Connected or synchronized data creates retention, deletion, residency, and compliance obligations.
- Third-party APIs can introduce separate costs, quotas, outages, and terms that OpenAI does not control.
- The platform and review requirements are evolving, so developers should budget for migrations.
- The framework is aimed at developers; it is not a drag-and-drop tool for nontechnical teams.
Compare
Apps SDK alternatives
The right alternative depends on the specific output, workflow, controls and budget your project requires.
Agents
Workspace Agents
Use OpenAI Workspace Agents when the main goal is a managed, shared team agent rather than a public service integration.
Explore Workspace Agents →Business Operations
n8n
Use n8n for self-hostable workflow automation across many services when an in-ChatGPT interface is not required.
Explore n8n →Project Management
MindPal
Use MindPal for a more visual, no-code approach to assembling multi-step AI workflows.
Explore MindPal →Agents
Zapier Agents
Use Zapier Agents when broad prebuilt app connectivity matters more than owning a custom MCP server and UI.
Explore Zapier Agents →Questions
Apps SDK FAQs
Is OpenAI Apps SDK still active?
Yes, but OpenAI now presents the capability under Plugins. The former Apps SDK documentation routes developers to the current Plugin platform, where MCP connections and optional in-chat UI sit alongside reusable skills.
What can an Apps SDK integration do?
It can expose structured MCP tools for finding information or requesting actions and can attach an interactive component when the result needs controls, visualization, or a richer layout.
Is the Apps SDK free?
OpenAI does not list a separate SDK fee. Developers still pay for their own infrastructure and third-party services, while end-user access depends on the relevant ChatGPT plan and workspace settings.
Do I need to use OpenAI models in my backend?
Not necessarily. The integration contract is the MCP server and package behavior. Your backend can use existing business logic and services, though separate model or API calls may add their own costs.
Can I use my existing customer accounts?
Yes. The platform supports authentication so the integration can respect existing accounts, permissions, and entitlements. The implementation must still minimize scopes and handle revocation safely.
Can a plugin change external data?
It can expose write actions, but consequential actions should show the user what will happen and require appropriate confirmation. Developers also need authorization checks, idempotency, and auditable error handling.
How do I publish an Apps SDK project?
Follow the current Plugin workflow: build and secure the MCP service, add optional UI or skills, package the required metadata, test the installed experience, and submit it through OpenAI's publishing process.
Is Apps SDK the same as the OpenAI Agents SDK?
No. Apps SDK, now part of Plugins, connects external tools, data, and optional UI to OpenAI clients. The Agents SDK is a separate framework for orchestrating model-driven agents in your own application.
Bottom line
Our Apps SDK verdict
OpenAI's Apps SDK remains a strong route for established products that want their data and workflows inside ChatGPT, but the name and distribution model have evolved. Build against the current Plugin documentation, start with narrow read-heavy tools, and treat authorization, confirmations, privacy, and operational reliability as core product work rather than submission cleanup.
Visit Apps SDK website ↗Similar Consumer Tools
View all Consumer AI tools →
DeepSeek V4.1 Flash
DeepSeek's, small, efficient new open model
ChatGPT Atlas
Atlas - OpenAI's new web browser built with ChatGPT at its core
GPT-6 Astra
OpenAI's next-generation frontier model
Scribe v2 Realtime
Scribe v2 Realtime- ElevenLabs' accurate, real-time transcription model with 150ms latency across 90+ languages
AI training for the future of work.
Get access to all our AI courses, hundreds of real-world AI use cases, live expert-led workshops, an exclusive network of AI early adopters, and more.
AI Courses
Get unlimited access to all of our current & upcoming industry-specific AI courses for the duration of your subscription.
Daily Guides
To keep up with the rapid pace of AI, our team publishes AI implementation guides daily. Our library contains 300+ practical use cases to automate real-world work.
Workshops
Join weekly, live, interactive sessions with industry leaders who are at the forefront of AI for hands-on implementation guidance and exclusive insights.
Community
Network with an exclusive community of AI-first professionals who are working smarter with AI. Learn how early adopters are using AI in their work and businesses.