--- type: "post" id: "openai-open-sources-mcp-extensions-for-chatgpt-native-672594" slug: "openai-open-sources-mcp-extensions-for-chatgpt-native-672594" title: "OpenAI open-sources MCP Extensions for ChatGPT-native plugins" description: "OpenAI has released MCP Extensions, an open-source SDK and specification for adding ChatGPT-specific interface capabilities to MCP servers and apps." retrieval_nugget: "OpenAI has released MCP Extensions, an open-source SDK and specification for adding ChatGPT-specific interface capabilities to MCP servers and apps." published_at: "2026-11-12" updated_at: "2026-11-12" record_date: "2026-11-12" date_kind: "published_at" topics: ["mcp","codex"] entities: ["OpenAI","MCP Extensions","ChatGPT","Model Context Protocol","Codex","Bits & Bolts"] related_patterns: ["agent-ready-software-exposes-capabilities"] novelty_label: "same_event" what_changed: null prior_refs: ["https://newruntime.com/posts/build-richer-more-interactive-plugins-for-chatgpt-analysis-73d222/"] source_url: "https://github.com/openai/mcp-extensions" source_urls: ["https://github.com/openai/mcp-extensions"] schema_version: "newruntime-agent-readable-v0.2" stable_id: "post:openai-open-sources-mcp-extensions-for-chatgpt-native-672594" status: "published" visuals: [] editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]} routes: {"html":"https://newruntime.com/posts/openai-open-sources-mcp-extensions-for-chatgpt-native-672594/","markdown":"https://newruntime.com/posts/openai-open-sources-mcp-extensions-for-chatgpt-native-672594.md","json":"https://newruntime.com/posts/openai-open-sources-mcp-extensions-for-chatgpt-native-672594.json"} markdown_url: "https://newruntime.com/posts/openai-open-sources-mcp-extensions-for-chatgpt-native-672594.md" --- # OpenAI open-sources MCP Extensions for ChatGPT-native plugins ## Retrieval answer OpenAI has released MCP Extensions, an open-source SDK and specification for adding ChatGPT-specific interface capabilities to MCP servers and apps. ## Summary OpenAI has released MCP Extensions, an open-source SDK and specification for adding ChatGPT-specific interface capabilities to MCP servers and apps. The repository documents sidebar entry points, custom file viewers, composer-based resource search and extended forms. For builders, the release offers a concrete path to make an MCP integration feel more like a native ChatGPT feature, rather than a generic tool endpoint. ## What happened OpenAI’s public GitHub repository describes MCP Extensions as a way to add ChatGPT-specific capabilities to MCP so developers can build plugins that feel like “native, first-class features.” The repository is public and carries an Apache-2.0 license. Its README presents several interface patterns: a plugin can provide a sidebar entry point so users can access an app from ChatGPT’s sidebar; a file extension handler can render a custom viewer when a user opens a supported file type; composer mentions can let users search a plugin’s resources and add references to a message; and extended forms can present thumbnail choices, with a CAD part picker given as the example. The project offers a TypeScript package, `@openai/mcp-extensions`, for MCP servers and Apps, and a Python package, `openai-mcp-extensions`, for MCP servers. The README directs developers to `docs/spec.md` for the supported extensions, `docs/patterns.md` for best practices and `CHANGELOG.md` for API changes. It also points to the Bits & Bolts source as a practical integration example. The repository’s showcased Bits & Bolts demo includes a Parts Library accessed through its sidebar icon and a walkthrough for exploring each major extension. OpenAI also describes a suggested starting workflow: ask “@Plugin Creator” in Codex to build a plugin or improve an existing one, install the relevant SDK, and consult the documentation and example. The README invites developers to submit feedback about gaps in the extensions. The repository page displayed 1.2k stars, 55 forks, nine issues, three pull requests and 30 commits when the page was captured; these are repository-page counts, not measures of production use. The source identifies both package languages and their stated use, but does not provide release dates, a pricing model, supported ChatGPT plans or a compatibility matrix. The important boundary is that the repository explicitly frames these as ChatGPT-specific capabilities added to MCP. It does not describe them as a universal MCP interface standard. A developer building for another MCP client should not assume that sidebar entry points, custom file viewers or composer interactions will transfer unchanged. The specification and changelog are the places the project points readers to for the extension surface and API changes. ## What the archive adds New Runtime’s earlier coverage described OpenAI plugins as moving beyond the conversation stream toward richer interfaces. The current release adds an SDK and documented extension surface to that thread: [Build richer, more interactive plugins for ChatGPT](https://newruntime.com/posts/build-richer-more-interactive-plugins-for-chatgpt-analysis-73d222/) (October 8, 2026). This is a new developer-facing release on a known product direction, not merely a restatement of the earlier interface analysis. ## Trend signal This supports the pattern of [agent-ready software exposing capabilities](https://newruntime.com/patterns/agent-ready-software-exposes-capabilities/): a product boundary becomes more useful when external agents and their users can reach specific, structured capabilities. The repository makes some user-facing plugin interactions explicit, including file handling and resource search. It does not establish how permissions, reliability or adoption work in practice, so the signal is about the exposed interface surface, not proof of a mature operating model. ## Why it matters MCP can describe a connection to tools, but a useful integration may need more than callable operations. A file viewer, a discoverable sidebar entry or a way to select a resource in the composer can make an integration fit into a user’s existing workflow. OpenAI’s SDK gives developers a stated implementation path for those interactions within ChatGPT, and a spec plus examples can reduce the need to infer an interface from a demo alone. The trade-off is platform specificity. These extensions are explicitly ChatGPT-specific, so teams should treat them as an additional client integration rather than assuming that a single MCP implementation automatically delivers the same experience everywhere. That may be worthwhile when ChatGPT is an important surface for a product; it can also mean extra implementation and maintenance work as APIs change. The repository points to its changelog for API changes, making version review part of adopting the SDK. For teams, the practical question is not whether every plugin needs a richer interface. It is whether a task benefits from being discoverable, previewable or selectable inside ChatGPT. A custom viewer may help when users open a supported file type; composer search may help when relevant plugin resources need to be referenced in a prompt. Those are examples the repository states, not measured claims about user outcomes. Teams should test the interaction against their own workflow before investing in a broader conversion. ## Limits and uncertainty The repository describes intended capabilities and supplies SDKs, documentation and an example, but the announcement does not establish how widely the extensions are available in ChatGPT, which plans support them or whether availability varies by account. It reports no independent evaluation of usability, reliability, security or developer effort, and gives no adoption figures for deployed plugins. The repository’s star and fork counts are not evidence of production usage. The source also does not state a support or compatibility policy across versions of ChatGPT or other MCP clients; developers should verify current behavior against the spec and changelog before relying on a particular extension. ## Skill to build now Choose one interaction in an existing MCP integration that is awkward as a plain tool call: for example, opening a supported file or finding a resource to cite. Review the `docs/spec.md` and Bits & Bolts example, implement that one interaction with the appropriate SDK, then test it in the ChatGPT workflow your users actually use. Check the changelog before shipping and keep the generic MCP functionality distinct from the ChatGPT-specific layer. ## Reader takeaway OpenAI’s MCP Extensions make several ChatGPT-specific plugin interactions available through TypeScript and Python SDKs, with a spec and demo to learn from. The opportunity is a more integrated experience in ChatGPT; the constraint is that these interface additions are not presented as portable features across MCP clients. Start with one interaction users need and validate it in context. Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json ## Continue in the graph Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`. - topic hub: [Mcp](https://newruntime.com/topics/mcp.md): Explore the mcp topic hub. - topic hub: [Codex](https://newruntime.com/topics/codex.md): Explore the codex topic hub. - collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis. - collection: [Sources](https://newruntime.com/sources/): See the original sources this archive cites. - whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)