VMGM Software
Start here

MCP server development

Your product, inside Claude and ChatGPT.

We build the MCP server that lets your customers use your product from the assistant they already have open, run it in your cloud, and get it listed in both directories.

Get a fixed quote

In short

  • VMGM Software builds MCP servers that let customers use a company’s product from Claude, ChatGPT and other AI assistants.
  • A VMGM Software MCP server is built on the client’s existing API, signs users in with OAuth and runs in the client’s own cloud account.
  • VMGM Software submits each MCP server to the Claude Connectors Directory, the ChatGPT Plugin Directory and the official MCP Registry, under the client’s company name.
  • VMGM Software built and runs the MCP server for View Printer, its own social media scheduler, which is listed in the MCP Registry and in Claude’s community plugin marketplace.
  • Every VMGM Software project is one fixed quote, free, and the client owns the code, the hosting and the listings.

What we build

An MCP server is how an AI assistant uses your product: a set of tools it can call, each one an action your customers take today by clicking through your app. We build that server on top of your existing API, and everything the directories expect around it:

  • Tools shaped around the jobs, not one per endpoint, each with a description written for the model that has to choose it
  • Sign-in with OAuth, so the assistant acts as the customer, with their permissions and nothing more
  • Every tool labeled read-only or write, and destructive where a change cannot be undone, so the assistant can ask before it acts
  • A log of every call, in your own monitoring
  • A skill, where the product has a workflow the assistant needs teaching, shipped with the server as a plugin

Connector, app, plugin: each assistant has its own name for this. Here is what each one means.

Where your product gets listed

  • The Claude Connectors Directory, for Claude users on the web, desktop and mobile, in Claude Code and in Cowork
  • The ChatGPT Plugin Directory, for ChatGPT and Codex users
  • The official MCP Registry, which MCP clients and smaller directories read from
  • Any assistant that accepts a server URL, such as Cursor, or your customers' own agents

One server covers all four. We submit it to each directory under your company's name, answer the reviewers, and fix what they ask for.

Meta's Muse opened its own connector directory in September 2026. Meta reviews each application for function, security and legal requirements, and we take clients through it alongside the other two.

How a project runs

  1. A call about what your customers would ask for. The list of things people would say to an assistant decides the tools, not the list of endpoints.
  2. A fixed quote, free, a few days later.
  3. The server, built in your cloud, next to the product it calls.
  4. Tested in Claude and ChatGPT as a custom connector, on a populated test account, before either directory sees it.
  5. Submitted and taken through review in both directories.
  6. Handed over. Your team owns the code, the hosting and the listings.

Who owns the server and the listings?

You do. The code is in your repository and the server runs in your cloud account. Both directories list a product under the company that publishes it, and OpenAI verifies that company's identity and domain, so the listings are in your name from the start.

MCP servers we have built

Our own · 2026

View Printer

Schedules and publishes social posts to TikTok, Instagram, YouTube, X and Facebook, and you talk to it rather than fill in a form.

TanStack · Postgres · AWS

Frequently asked questions

Do we need an API before we can have an MCP server?

It helps, and most products have one, even if only their own app uses it. Where the server needs something the API does not offer, we add the endpoint as part of the project.

Is it safe to let an AI take actions in our product?

The assistant can only do what the signed-in customer could do, because it acts with their account through OAuth. Tools that change or delete something are labeled so the assistant can ask first, and every call is logged.

How long do the Claude and ChatGPT reviews take?

Neither publishes a timeline. Anthropic scans a submission automatically and lists it as a Community connector by default, with some submissions also reviewed by a person. OpenAI reviews every submission and says timelines vary.

Will it work in Cursor and other assistants?

Yes. MCP is an open standard, so any assistant that accepts a server URL can use it, including Cursor, Claude Code, Codex and agents your customers build themselves.

Can you get our product into Meta’s Muse?

Yes. Muse takes connectors by application, and today that is a single form at muse.ai/platform, followed by Meta’s review and testing. We have applied with View Printer, and we take clients through the same process alongside Claude and ChatGPT.

How much does an MCP server cost?

It depends on how many tools the product needs, how sign-in works today and whether the API needs new endpoints. Here is what moves the price, and the fixed quote is free.

Related

What are Claude connectors and ChatGPT plugins?

What MCP servers, connectors and plugins are, and what each directory asks for.

Tell us what you want built, and a fixed quote follows within days.

Start here