What is MCP, and why should sales teams care?

MCP, the Model Context Protocol, is an open standard that lets an AI assistant such as Claude or ChatGPT call the tools of another product through one connector. For a sales team it means you can ask your assistant about your pipeline, or tell it to act in your outreach tool, without copying data in and out. What matters is which actions the connector exposes and who approves them.

This post explains the idea without jargon, shows what it changes for a sales team, and gives a short list of questions to ask before you connect an assistant to anything that can send a message.

What problem does MCP solve?

Before MCP, connecting an assistant to a tool meant a custom integration for every pair. One for Claude and your CRM, another for ChatGPT and your CRM, another for each new assistant. Each one had its own way of logging in, its own list of actions and its own bugs.

MCP replaces that with one shared language. A product publishes an MCP server that lists what it can do, for example "list leads", "pause a campaign" or "show the conversation with this contact". An assistant that speaks MCP can read that list and use the actions. The protocol documentation describes it as a standard way to connect AI applications to tools and data, and Anthropic introduced it in November 2024. Other assistants have adopted it since, which is the point of a standard.

The easiest picture is a power socket. The tool has one socket shape. Any assistant with the matching plug can use it.

What does a connector look like to a user?

For the person using it there is no code. In Claude you open Settings, then Connectors, add a custom connector with a URL, sign in and approve. ChatGPT and other clients offer the same idea in their own connector settings. After that you write in plain language.

Our own connector is a good example because we can tell you exactly what it exposes. It has 43 tools and 4 saved workflows. The tools fall into groups: leads and contacts, prospecting, conversations across channels, campaigns, approvals, meetings, analytics and search. Sign in uses OAuth, you pick a workspace and an access level, Full access or Read only, and every request is logged.

So a question such as "which of my replied leads have not had an answer for two days" turns into a few tool calls, and the assistant shows you the list. A request such as "pause the campaign that targets agencies" turns into one more.

What changes for a sales team?

Three things change, and none of them is magic.

Reading gets faster. Most of a sales leader's week is questions: who replied, which campaign is slow, what happened with this account. An assistant with read access answers them from live data instead of from a stale export. The weekly pipeline review recipe and the stale lead audit show what this looks like in practice.

Routine changes get easier. Pausing a campaign, editing a first message or picking up the leads that need a person are small jobs that cost clicks. An assistant does them from one sentence.

Context travels with you. Instead of switching between a CRM, an inbox and an outreach tool, you ask in one place. The assistant sees the full conversation history across channels, so it can brief you before a meeting.

What does not change is the quality of the underlying work. An assistant with 43 tools is not a better list or a better offer. On our own LinkedIn outbound, the biggest differences came from the audience we picked and how the first message opened, not from how fast anything was operated. In that data, people who had just engaged with a post about sales automation replied at 54.6%, against 21.7% for a Sales Navigator search (Sellinger case study). A connector helps you run the work. It does not choose the audience for you.

What can go wrong?

An assistant that can act can also act wrongly. Four risks are worth naming.

1. Overreach. Giving write access to everyone when most people only need to read. Read only access exists for this reason.
2. Unreviewed sends. An assistant that sends messages in your name on a bad day. Look for a confirmation step, or for an approval queue where a person approves each message.
3. Wrong workspace. A connector that can see every account when you wanted one. Tokens should be scoped to one workspace.
4. No trail. If something goes wrong you need to know who asked for what. An audit log answers that.

None of these is a reason to avoid MCP. They are reasons to ask the vendor.

Which questions should you ask before connecting?

• Which actions does the connector expose, and can I see the full list?
• Is there a read only mode?
• Does anything that sends a message need my confirmation?
• Is the access scoped to one workspace and one level?
• Are requests logged, and can I end access at once?
• How long do sign in tokens last?

For Sellinger the answers are in the open: the list of tools is on the MCP page, tokens are scoped to one workspace and one access level, access tokens last 8 hours and refresh automatically for up to 90 days, requests are logged for security auditing, and access ends the moment you disconnect the connector or are removed from the workspace.

Is MCP the same as an API?

No, though they are related. An API is a set of endpoints a developer calls with code. An MCP server sits on top and describes those capabilities in a form an assistant can understand and choose from, with plain language names and descriptions. Sellinger also offers a REST API with webhooks, from the Full Stack bundle up, for people who want to write code. The connector is for people who would rather ask.

Should a sales team care now?

Care in proportion. If your team already uses Claude or ChatGPT every day, a connector to your pipeline removes a lot of copying and is worth an afternoon to try, with read only access first. If your outreach is small and your process is still settling, the audience, the offer and the opener matter much more than the interface.

A sensible way to start:

1. Connect with Read only access and ask five questions you ask every week.
2. Compare the answers with what you see in the app. Notice what it gets right and wrong.
3. Switch to Full access only for the one or two actions you do daily, and keep confirmations on.
4. Add a saved workflow for the review you run every Monday.

For ideas, the MCP recipes are a tested starting point, and the post on what ChatGPT and Claude can and cannot do for outreach goes through the limits in more detail. The glossary entry has the short definition.