Your platform directly in ChatGPT and Claude: What MCP means for software providers

Key takeaways

• The Model Context Protocol, MCP for short, is an open standard that lets AI assistants such as Claude and ChatGPT use data and functions from other systems. Anthropic introduced it in November 2024, and since December 2025 it has been part of the Agentic AI Foundation under the Linux Foundation.
• An MCP server describes the functions of your platform as tools that an AI model understands and calls on its own. Your customers then use your product directly in the chat they already work in.
• Claude connects MCP servers as connectors, ChatGPT as apps, which are now bundled into plugins. Both providers run directories where you can submit your server.
• A good MCP server offers a few clearly described tools, signs users in via OAuth where needed and delivers reliable data. Tools that complete an entire workflow in one call work particularly well.

As of August 2026

Your customer plans her week with Claude or ChatGPT. She wants to know which orders are still open in your software and add the next appointment right away. Until now, that means leaving the chat, opening your application and copying data across. With an MCP server, the assistant handles this directly in the conversation, using your platform's real data.

What is the Model Context Protocol?

The Model Context Protocol (MCP) is an open standard that connects AI applications to external systems. Anthropic, the company behind Claude, released it on November 25, 2024. The official documentation compares MCP to a USB-C port: just as USB-C connects devices uniformly, MCP uniformly connects AI applications to data, tools and workflows.

For providers, it matters that no single company owns the standard. In December 2025, Anthropic handed MCP over to the Agentic AI Foundation, a fund under the Linux Foundation co-founded by Anthropic, Block and OpenAI. According to the project maintainers, the main software development kits (SDKs) for MCP are downloaded close to half a billion times a month.

How an MCP server works

MCP has three roles. The host is the AI application, such as Claude or ChatGPT. Inside it, a client maintains the connection, and on the other side is the server you operate as the provider. According to the specification, your server provides three kinds of building blocks:

• Tools: functions that the AI model calls on its own, for example "create appointment" or "find invoice".
• Resources: data and content that the application passes to the model as context, such as a document.
• Prompts: prepared instructions that users select deliberately, for example from a menu.

In everyday use, tools are at the center. Each one has a name, a description and a schema, meaning a formal list of the expected inputs. The assistant asks your server which tools are available, picks one that fits the request, calls it and works with the result. The specification recommends that a human can always step in, for example through confirmation prompts before actions.

Which AI assistants support MCP

Since May 2025, Claude has connected remote MCP servers running on the internet, which Anthropic now calls connectors. A shared directory covers claude.ai, the desktop and mobile apps, Claude Code and Cowork, and Claude proactively suggests suitable connectors in conversations.

OpenAI introduced apps in ChatGPT on October 6, 2025. According to OpenAI, the accompanying Apps SDK builds on MCP, and at the time OpenAI said developers could reach more than 800 million ChatGPT users with it. OpenAI now bundles apps and skills, meaning reusable instructions, into plugins with a shared directory for ChatGPT and the developer tool Codex. OpenAI says availability depends on factors including the app, the plan and the region.

Anthropic also names Cursor, Gemini, Microsoft Copilot and Visual Studio Code as products that have adopted the standard, and counted more than 10,000 active public MCP servers in December 2025. Since January 2026, MCP Apps have also let a tool show its own interfaces such as forms or charts directly in the chat, supported by Claude and ChatGPT, among others.

What this means for SaaS and platform providers

Your product becomes usable where people already work with AI. Accounts, permissions and rules stay in your application, and the assistant becomes an additional channel to your customers.

Because MCP is open, one server lets you reach several assistants. The MCP documentation sums this up as "build once and integrate everywhere".

Good tools follow your customers' tasks

OpenAI advises developers to align every tool with a user goal instead of simply mirroring an internal interface. The description explains what the tool does and when it fits, because the model uses it to choose. Reading and changing functions belong in separate tools.

For Claude's directory, every tool needs a title and a label stating whether it only reads or changes data. These labels control confirmations: read-only tools can run without one, while changing tools always ask.

Entire workflows instead of many individual steps

More tools don't automatically lead to better results, Anthropic writes in its guide to tools. AI models can only process a limited amount of information at once. Anthropic therefore recommends combining frequently chained steps into a single tool. Its example: instead of three tools for listing people, listing appointments and creating an appointment, use one that finds a free slot and books it right away.

OpenAI recommends the same: operations that together form one action belong in one tool. They are split where permissions, risks or the required confirmation differ. This way your business rules run on your server instead of the model piecing them together from individual steps.

Sign-in via OAuth: customers use their existing account

If a tool accesses non-public data or acts on behalf of a user, it should sit behind a sign-in. For servers on the internet, the MCP specification builds on OAuth 2.1, the evolution of the widely used authorization standard OAuth 2.0. Your customer signs in with her existing account with you, and the assistant receives an access key valid only for your server and only for the permissions granted.

Anthropic requires OAuth for connectors in its directory whenever a service needs a sign-in. For plugins, OpenAI points to the same flow from the MCP specification. For public data, such as a product catalog, a server can also work without sign-in.

Reliable data and clear error messages

An assistant is only as good as the answers it gets from your server. Anthropic recommends returning only meaningful information, and for the Claude directory, responses should be reasonably sized for their purpose. Error messages explain what to do instead of just reporting "server error". The specification provides dedicated error results that let the model correct its inputs and retry, for example when a date must be in the future.

The path to your own MCP server

Start with two or three tasks your customers would most like to hand over to an assistant. If you already have an interface (API), the MCP server builds on it. For the directories, it runs at a fixed, encrypted internet address (HTTPS). You can test it beforehand in Claude as a custom connector and in ChatGPT as a custom MCP server.

You submit to Anthropic through a developer portal available on every paid Claude plan, and to OpenAI as a plugin through the OpenAI platform. Anthropic initially lists submissions as community connectors after an automated check, while at OpenAI you decide when your approved plugin goes live.

In practice

byclaw.io is our own product, an AI shopping agent that people use via WhatsApp, on the web or directly in Claude and ChatGPT. For the two assistants, we built an MCP server that handles search, comparison and recommendation in a single tool call. This gives the assistant a finished result in a single call that it can pass straight on to the user. It is exactly the approach Anthropic and OpenAI recommend: one tool for a complete goal instead of many individual steps.

Reliable data complete the picture. The catalog updates daily from affiliate feeds, meaning product lists from retailers' partner programs, and a hallucination guard only lets real, linkable products through.

Frequently asked questions

Is an MCP server the same as an API?
Not quite. An API is aimed at programmers who know exactly which function they are calling. An MCP server describes its tools so that an AI model recognizes on its own when it needs which one. Behind the scenes, it usually uses your existing API.

Do I need two different servers for Claude and ChatGPT?
No. Both assistants connect MCP servers reachable over HTTPS and follow the same open standard. The differences lie in packaging and review: Anthropic lists servers as connectors, OpenAI as part of a plugin. So you test and submit with both.

How do users find my platform in the assistant?
Existing customers add your server via its address or install it from the directory. Claude suggests directory connectors in the conversation when they fit the task, and ChatGPT can also suggest suitable apps. This makes your offer visible to people who don't know your platform yet.

Can AI agents also buy from me via MCP?
Purchases and payments by AI agents are a separate topic with their own open standards, such as the Agentic Commerce Protocol, which OpenAI and Stripe developed, initially for purchases directly in ChatGPT. These standards cover the cart, checkout and orders, among other things.

Conclusion

With one open standard, MCP makes your platform usable where your customers already work with AI, in Claude, ChatGPT and other applications. It takes a few clearly described tools, sign-in via OAuth and reliable data. Tools that complete an entire workflow on your server work particularly well. An existing interface is a good starting point, and your platform gains a new channel to customers.

Your platform in ChatGPT and Claude?

Describe your project in our AI project estimator. You will receive a free estimate by email, and we will get back to you within 24 hours.

Learn more about our services for SaaS products and platforms and marketplaces. Or simply write to us at our email address.

Sources

1. Anthropic: "Introducing the Model Context Protocol", November 25, 2024. https://www.anthropic.com/news/model-context-protocol
2. Anthropic: "Donating the Model Context Protocol and establishing the Agentic AI Foundation", December 9, 2025. https://www.anthropic.com/news/donating-the-model-context-protocol-and-establishing-of-the-agentic-ai-foundation
3. Model Context Protocol: "What is the Model Context Protocol (MCP)?". https://modelcontextprotocol.io/docs/getting-started/intro
4. Model Context Protocol: specification, version 2026-07-28, sections Overview, Server Features and Tools. https://modelcontextprotocol.io/specification/2026-07-28, https://modelcontextprotocol.io/specification/2026-07-28/server, https://modelcontextprotocol.io/specification/2026-07-28/server/tools
5. Model Context Protocol: specification, version 2026-07-28, section "Authorization". https://modelcontextprotocol.io/specification/2026-07-28/basic/authorization
6. Model Context Protocol Blog: "The 2026-07-28 Specification", July 28, 2026. https://blog.modelcontextprotocol.io/posts/2026-07-28/
7. Model Context Protocol Blog: "MCP Apps - Bringing UI Capabilities To MCP Clients", January 26, 2026. https://blog.modelcontextprotocol.io/posts/2026-01-26-mcp-apps/
8. Anthropic: "Claude can now connect to your world", May 1, 2025, updated June 3, 2025. https://claude.com/blog/integrations
9. Claude Help Center: "Get started with custom connectors using remote MCP". https://support.claude.com/en/articles/11175166-get-started-with-custom-connectors-using-remote-mcp
10. Claude Docs: "Connectors directory", "Submit a connector to the directory" and "Connector pre-submission checklist". https://claude.com/docs/connectors/directory, https://claude.com/docs/connectors/building/submission, https://claude.com/docs/connectors/building/review-criteria
11. Anthropic Engineering: "Writing effective tools for agents — with agents", September 11, 2025. https://www.anthropic.com/engineering/writing-tools-for-agents
12. OpenAI: "Introducing apps in ChatGPT and the new Apps SDK", October 6, 2025, updated November 13, 2025. https://openai.com/index/introducing-apps-in-chatgpt/
13. OpenAI: "Buy it in ChatGPT: Instant Checkout and the Agentic Commerce Protocol", September 29, 2025. https://openai.com/index/buy-it-in-chatgpt/
14. OpenAI Help Center: "Connected apps in ChatGPT". https://help.openai.com/en/articles/11487775-apps-in-chatgpt
15. OpenAI Help Center: "Plugins in ChatGPT". https://help.openai.com/en/articles/20001256-plugins-in-chatgpt-and-codex
16. OpenAI Developers: Plugins documentation, pages "Plugin architecture", "MCP server", "Define tools", "Connect and test your plugin" and "Upload and submit your plugin". https://developers.openai.com/plugins