MCP and tool use

Tool use is the model calling a function instead of guessing: look this up, run this query, create this ticket. MCP (Model Context Protocol) is the standard plug that lets any assistant connect to any such tool, so the same connector for Linear, Notion or a database works in Claude, ChatGPT and the rest. A connected assistant reads real data and acts in real systems.

Standing: Established since 2025-06
Stars, week: +3,855 across 6 repositories
Stars, month: +12,331, through 2026-11-08
Listings: 6 chosen by hand

The standing

Where it stands

Among the 14 concepts. Bars: GitHub stars gained this week.

Established · 7

Rising · 7

How to use it today

In ChatGPT or Claude, as a person doing the work, not as a developer.

1In Claude open Connectors; in ChatGPT open the Plugin Directory. Add the apps you work in (Google Drive, Slack, Notion, Linear, GitHub); each one is an MCP server or works like one.
2Ask questions that need live data ("what did customers say in support tickets this week") and watch which tool the assistant calls; correct it if it reached for the wrong one.
3For tools without an official connector, pick a listed MCP server from this catalog and paste its setup line into your app; prefer official or reviewed servers.
4Give the assistant read access first; grant write access (creating tickets, sending messages) only for actions you would delegate to a new colleague.
5Keep a note of which servers you connected and what each one can do; a connector is a permission.

What changes day to day

You stop copying data into the chat. Questions get asked against the systems themselves, and the assistant's answers carry references you can open. Some chats end with an action taken rather than text to paste.

Where it goes wrong

A third party server sees whatever the assistant sends it; an unreviewed one can exfiltrate data or follow instructions planted in a document. Too many connected tools make the assistant slower and less accurate at choosing among them.

Listings that use it well

Chosen by hand after reading them, each with the reason. Install from the listing's own page.

Listing	Kind	Why it is here	Stars
mcp-builder	Agent skills	Anthropic's guide to designing MCP servers with tools an LLM can use well.	190k
composio	Agent skills	1000+ toolkits with tool search and auth so an agent can act in Gmail, Slack, GitHub or Notion through the SDK or MCP.	31k
fastmcp	Agent skills	The Pythonic framework for building MCP servers and clients; its skill lists and calls tools on any server.	28k
GitHub	MCP server	GitHub's official server: repos, issues, PRs and workflows become callable tools.	34k
mcp-for-beginners	MCP server	Microsoft's curriculum teaching MCP fundamentals with examples in six languages.	18k
typescript-sdk	MCP server	The official MCP TypeScript SDK for servers and clients.	14k

Elsewhere

Repositories that embody the idea and are not in the catalog.

modelcontextprotocol/servers ↗Reference MCP servers (filesystem, git, fetch, memory, time) maintained by the protocol project.
modelcontextprotocol/modelcontextprotocol ↗The specification, schema and documentation source.
punkpeye/awesome-mcp-servers ↗The most used community catalog of MCP servers.

Standing

Our judgment, with the reason and the sources it rests on.

Established since 2025-06. Introduced in late 2024, adopted by OpenAI in March 2025 and by every major vendor during 2025, and governed by the Linux Foundation's Agentic AI Foundation since December 2025. The July 2026 revision made the protocol stateless and added official extensions, including MCP Apps, which show a live interface inside the chat. The question is which servers to trust, not whether to use it. Settled practice: the methods are known, the tools are mature, and the question is how well you use it.

OpenAI Developers on X: MCP and the OpenAI Agents SDK2025-03-26
MCP joins the Agentic AI Foundation2025-12-09
The 2026-07-28 Specification2026-07-28