Who owns the Model Context Protocol?

Anthropic introduced MCP in November 2024 and donated it in December 2025 to the Agentic AI Foundation, part of the Linux Foundation. It is developed in the open as a shared standard.

What is the current MCP specification version?

The current version is 2026-07-28, released on July 28, 2026. It makes the core protocol stateless and officially deprecates the older HTTP+SSE transport.

Should I choose a local stdio server or a remote Streamable HTTP server?

A local stdio server keeps data on your machine but runs with your user permissions, so run it in a sandbox or container. A remote server is easier to share across a team, but your data passes through the host's infrastructure, so check its authentication, scopes and logging.

Can AI agents browse the GreenGo catalog?

Yes. GreenGo publishes a read-only MCP endpoint at the /mcp path so agents can browse listings.

What is an MCP server, and how do you buy or sell one?

Short answer: An MCP server is a small program that gives AI apps like Claude, ChatGPT or Cursor a standard way to use your tools and data, through the open Model Context Protocol. You buy one the way you'd buy any software: as code you run yourself or as a hosted endpoint. Before you pay, check how it handles logins, permissions, hosting and updates.

What the Model Context Protocol is

The Model Context Protocol (MCP) is an open standard for connecting AI applications to outside systems. It was created at Anthropic by David Soria Parra and Justin Spahr-Summers and introduced on November 25, 2024. Other major AI providers adopted it later, including OpenAI in March 2025.

In December 2025, Anthropic donated MCP to the Agentic AI Foundation, a new foundation under the Linux Foundation that was announced on December 9, 2025. The protocol is now developed in the open as a Linux Foundation project.

The current specification is version 2026-07-28, released July 28, 2026. Its biggest change is that MCP now works as a stateless request/response protocol. Each request carries its own protocol version and capabilities, and the old initialize handshake and Mcp-Session-Id header are retired. Servers become easier to run behind ordinary load balancers.

How an MCP server talks to an AI app

An AI app (the client) connects to an MCP server over a transport. The spec defines two standard transports:

stdio: the client starts the server as a local subprocess and they exchange newline-delimited messages over standard input and output. This suits tools that run on your own computer.
Streamable HTTP: each message is an HTTP POST to a single MCP endpoint, and replies come back as JSON or a short-lived event stream. This suits hosted, remote servers. In the 2026-07-28 version, these requests must include Mcp-Method and Mcp-Name headers.

The older HTTP+SSE transport is now officially deprecated, with a year-long transition period. If a server you're considering only supports HTTP+SSE, plan for an update.

What a server exposes

An MCP server can offer three kinds of building blocks:

Primitive	Who controls it	What it is	Example
Tools	The model	Functions the AI can call to take actions	Create an invoice, query a database
Resources	The application	Data the client attaches as context	File contents, git history
Prompts	The user	Reusable templates the user picks	Slash commands, menu options

Most commercial servers focus on tools. A good listing tells you exactly which tools exist and what each one can change.

Where MCP servers are listed

The official MCP Registry launched in preview on September 8, 2025, at registry.modelcontextprotocol.io. It's an open catalog and API built from information that server maintainers publish themselves, and other directories can build on it.
The GitHub MCP Registry launched on September 16, 2025, as a place to find servers for Copilot and other MCP clients.
Marketplaces like GreenGo's MCP servers catalog list servers you can buy, including paid code and hosted access. GreenGo also publishes a read-only MCP endpoint at /mcp, so an agent can browse the catalog directly.

How to evaluate an MCP server before you buy

Use this checklist on any listing:

Authentication. For remote servers, look for standard OAuth. The MCP spec says servers must not accept tokens that weren't issued for them, so a server that simply passes your token through to another API is a warning sign.
Scopes and permissions. Ask for the smallest set of permissions that does the job. The MCP security guidance warns against wildcard scopes like *, all or full-access. Read-only tools are easier to trust than tools that write or delete.
Hosting. A local stdio server runs with your user permissions on your computer. Read the exact launch command, and prefer running it in a container or sandbox. A hosted server means your data passes through someone else's infrastructure, so ask where it runs and what it logs.
Spec version. Check whether it supports 2026-07-28, an earlier version, or both. Roots, Sampling and Logging are deprecated in the new version but keep working for at least 12 months.
Maintenance. Look for a changelog, recent releases and a clear support channel. MCP is still changing, and an abandoned server will fall behind.
License. Know whether you can use it commercially, change the code, or deploy it for clients.

For a broader look at agents that use MCP servers, read AI agents you can buy today.

How sellers package an MCP server

If you built an MCP server with Claude Code, Cursor or another AI tool, buyers will want a clean, documented package:

A README that lists every tool, the scopes it needs, its environment variables and the spec versions it supports.
A clear delivery format: source code, a package or container image, or access to a hosted endpoint.
No secrets in the files. Ship an .env.example file, never a real .env.
A changelog and update policy, so buyers know what they get after purchase.

Pricing models that fit MCP servers

Model	Fits when	What the buyer gets
One-time license	Buyer self-hosts the code	Source code plus updates for a set period
Tiered license	Personal vs. commercial or agency use	Same code, different usage rights
Subscription	You host and maintain the endpoint	Ongoing access, uptime and updates
Pay-per-use	Usage varies a lot between buyers	Metered calls or tasks

Pricing logic for AI-built software is covered in detail in From prototype to paid.

How to list an MCP server on GreenGo

Go to Sell and create a listing. Listing is free.
Choose the MCP server category and describe the tools, scopes, transports and supported spec versions.
Upload your delivery files or connect your hosted endpoint, and run the automated Green Light checks. They look at things like DNS ownership, security headers, exposed secrets, known vulnerable dependencies and stray files such as .env or source maps. A pass is shown for 90 days with the check date. It is not a security audit or certification.
Pick a plan on Pricing: Starter takes 8% per sale, and Pro is $29/month plus 4% per sale. Buyers pay no extra fees, and paid purchases include 14-day Buyer Protection.

Sources

Updated October 20, 2026.