What is MCP? A plain guide for agencies
Published
Short answer
MCP, the Model Context Protocol, is an open standard that lets AI assistants like Claude, ChatGPT, Codex and Cursor use other software on your behalf. A tool that offers an MCP server describes the actions it supports, such as "create a contract" or "log time", and the assistant calls them when you ask. You sign in once and decide what it may do.
MCP in one paragraph
An AI assistant on its own can only talk. It can write a contract, but it can't send it, save it or check whether it's been signed. MCP is the plug that connects the assistant to the tools where that work happens. Anthropic introduced it as an open standard, and it's now supported by Claude, ChatGPT, Codex, Cursor and many others. One connection works in every assistant that speaks MCP.
How it works, without the jargon
- The tool publishes a menu. An MCP server lists the actions it offers: "list contracts", "create a contract from a template", "send for signature", and what each needs.
- You connect and sign in. The assistant opens the tool's sign-in page. You log in and choose what it may do, for example read only or read and write.
- You ask in plain words. "Send the NDA for Fjord Studio to Jane."
- The assistant picks the actions. It finds the client, creates the contract from your NDA template and asks you to confirm before sending.
The assistant doesn't get your password, and the tool doesn't get your whole conversation, only the actions the assistant asks it to take.
The connection between a request and an action
Your AI assistant
Turns your request into calls to the tool’s available actions.
Pledge workspace
Runs the permitted action and returns its result.
You choose the workspace and access. The tool still enforces its permissions.
MCP versus an API or Zapier
| API | Zapier-style automation | MCP | |
|---|---|---|---|
| Who uses it | Developers | Anyone, with set-up | Anyone, in plain words |
| Set-up | Code | A fixed workflow per task | Connect once |
| Handles new requests | Only what's coded | Only the workflow you built | Anything the tool's actions allow |
| Where you work | Your own app | Background | Your AI assistant |
What MCP means for an agency
Most agency admin is the same handful of tasks: turning a yes into a contract, chasing signatures, logging time, checking who's free. With MCP those happen in the chat where you're already working. For contracts that means:
- Drafting from your own templates instead of generic text
- Sending for signature from the conversation
- Asking "what's still unsigned?" instead of opening another app
What to check before you connect a tool
- Permissions follow you. The assistant should act as you, with your role, not as an admin.
- Read-only is an option. Useful for reporting and research.
- Destructive actions are limited. Pledge, for example, has no delete tools at all.
- Important steps need you. Sending a contract should never happen without your say-so.
- There's a trail. Changes made through the assistant should be logged.
See is it safe to let an AI agent send contracts? for more.
Questions
- What does MCP stand for?
- Model Context Protocol. It's an open standard for connecting AI assistants to other software and data.
- Which AI assistants support MCP?
- Claude, ChatGPT, Codex and Cursor all support MCP servers, along with a growing number of other assistants and developer tools.
- Do I need to be technical to use MCP?
- No. In most assistants you add a connector or app, sign in, and approve access. No code is involved.
- Is MCP secure?
- MCP itself is a protocol. Security depends on how each tool implements it: look for sign-in with your own account, permissions that follow your role, a read-only option and a log of changes.
Try Pledge
Contracts, time, plans and leads for small agencies, with an MCP server for Claude and ChatGPT. Pledge is in private beta: leave your email and we will send an invite.