Glossary
The terms this site uses, each defined in a sentence or two from the specifications and platform documentation we cite.
- Agent Distribution Engineering
- Agent Distribution Engineering is the work of making an existing SaaS product listed, working and chosen inside AI agents: tool design, auth, evals, directory submission and measurement after launch.
- What Agent Distribution Engineering covers
- Model Context Protocol (MCP)
- An open protocol that lets AI applications connect to outside tools and data. An MCP server offers tools, resources and prompts; an AI application such as Claude, ChatGPT or Cursor connects to it as an MCP client. The current revision is 2026-07-28.
- MCP server
- A program that offers tools, resources or prompts to AI applications over MCP. A local server runs on the user's machine and talks over stdio; a remote server runs on the web, talks over Streamable HTTP and serves many users.
- Connector
- Claude's name for a remote MCP server listed in its Connectors Directory. You submit it as a server URL, and Claude connects to it on each user's behalf.
- Claude Connectors Directory requirements
- Plugin
- A packaged extension for an AI product. In ChatGPT, an MCP-backed app listed in the Plugins Directory that ChatGPT and Codex share; in Claude, a bundle hosted on GitHub; in Cursor, a Git repository with a plugin manifest.
- Claude Connectors Directory
- Anthropic's directory of MCP connectors. Submissions go through the directory submission portal, automated checks and a safety scan. New listings carry the Community label, and Anthropic may escalate a connector to Verified review on its own.
- Every requirement, with sources
- ChatGPT Plugins Directory
- OpenAI's directory of plugins, shared by ChatGPT and Codex. A submission needs a verified identity, a domain verification token, exactly five positive and three negative test cases, a demo recording and release notes.
- Every requirement, with sources
- Cursor Marketplace
- Cursor's directory of plugins. Plugins are submitted from a public Git repository at cursor.com/marketplace/publish and reviewed by hand before they are listed.
- Every requirement, with sources
- Official MCP Registry
- The MCP project's public registry of MCP servers, still in preview. Each entry is published from a server.json, under a namespace the publisher authenticates through GitHub, DNS or HTTP.
- Every requirement, with sources
- server.json
- The file that describes an MCP server to the Official MCP Registry: its name, its version and, for a remote server, the URL listed in its remotes property.
- The registry's server.json rule
- Tool annotations
- Hints a tool declares about its behavior: readOnlyHint, destructiveHint, idempotentHint and openWorldHint. The MCP spec tells clients to treat them as untrusted unless the server is trusted, and Claude's and ChatGPT's reviewers check them.
- Claude's annotation rule
- server/discover
- The request every MCP server must answer under the 2026-07-28 revision, giving the protocol versions, capabilities and identity it supports. That revision removed the initialize handshake, which older clients still send.
- Moving a server to the current spec
- Client ID Metadata Document
- An HTTPS URL that an MCP client uses as its OAuth client ID. The URL serves a JSON document describing the client, which the authorization server fetches and checks, so the two need no prior registration.
- Dynamic Client Registration (DCR)
- An OAuth standard (RFC 7591) that lets a client register itself with an authorization server. MCP 2026-07-28 deprecates it in favor of Client ID Metadata Documents and keeps it for backward compatibility.
- Evals
- Automated tests that give an AI agent connected to your server realistic prompts and check that it picks the right tool and gets a correct result.