01 / The MCP Playbook for GTM Teams

Your AI, finally connected to your real work.

MCP is the open standard that lets Claude, ChatGPT, and Cursor reach into the tools your team uses every day — your CRM, your content, your inbox, your tickets. Here's what it is, what it changes for your team, and how to get it set up this week.

22 min readUpdated April 2026By Ben Ard, MassetFor: Sales · Marketing · CS · RevOps · Enablement
AI Client
Claude
The AI tool your team uses
connected
Connect
MCP Server
Masset
Your content library, live
ready
connection log
→ AIConnectAuthorize Claude to use Masset
02 / What is MCP

Think of MCP as USB-C for AI.

Before USB-C, every device shipped with its own cable. Phones, cameras, printers, hard drives — each one a different port, a different driver, a different drawer of tangled wires. USB-C ended that. One cable, one shape, one protocol. Plug it in and it works.

MCP — the Model Context Protocol — does the same thing for AI. It's an open standard, originally introduced by Anthropic in late 2024 and now used by Claude, ChatGPT, Cursor, Windsurf, Gemini, and most other major clients. Build one MCP server for your product, and it works inside every one of them.

The animation below is the flat-pack version. One AI client in the middle. Six MCP servers around it. Every spoke is a tool the model can call — search a doc, send an email, query a database, file a ticket — without anyone wiring up a custom integration.

Masset
content
Gmail
email
HubSpot
CRM
Slack
messaging
GitHub
code
Linear
tickets
03 / Why it matters

Without MCP, your AI is guessing.

Most AI tools today are stuck between two bad options. Either the model invents an answer based on its training data — confidently wrong, often subtly off — or you spend an afternoon copy-pasting context from five tabs into the chat just to get a useful first draft.

MCP eliminates both. The model still does the thinking, but the source of truth lives in your systems: your CRM, your content library, your ticket tracker, your documents. The model reaches into them in the moment, with your permission, and the answer it produces is grounded in your actual data.

without mcp
"Generic, polished, slightly off."

Sales rep asks AI for the latest healthcare case study. Model invents a company name and stats that look plausible. Email goes out. Prospect Googles. Trust gone.

with mcp
"Specific, sourced, sendable."

Same prompt — but the AI now searches your real content library, retrieves the actual "Memorial Health" study with real stats, and drafts the email with a tracked share link. Done in seconds.

04 / How it works

One question. The right answer in seconds.

When someone on your team asks an AI for something, MCP is what lets it actually go do the work — search your library, pull a deal record, draft an email. The animation below walks through a real two-tool sequence: an AE asks for a case study, Claude pulls it from Masset, and drafts the follow-up in Gmail. No copy-paste between tabs. No reps making up stats.

What your AI gets to do

Every MCP-connected tool gives your AI three kinds of superpowers. Once you understand them, you can predict exactly what each new MCP you install will let your team do.

01 / capability
Search

Find anything — the right deck, the right deal, the right ticket — across your real systems. Not generic web results. Your data.

02 / capability
Retrieve

Pull the actual file, record, or page into the conversation. The AI is reading your real data — not making something up that sounds about right.

03 / capability
Act

Draft an email. Update a deal. File a ticket. Post to Slack. Whatever the connected tool can do, your AI can now do for you.

05 / Install in 5 minutes

Send this section to your IT or RevOps lead.

Adding an MCP is a one-time, copy-paste setup. The exact steps for the four AI tools your team is most likely using are below. If you're not the one installing it, screenshot this section and send it to whoever handles tools at your company — they'll have it done before lunch.

claude_desktop_config.json
{
  "mcpServers": {
    "<server-name>": {
      "command": "npx",
      "args": ["-y", "<vendor-package>"],
      "env": {
        "API_KEY": "<your-api-key>"
      }
    }
  }
}
macOS: ~/Library/Application Support/Claude/ · Windows: %APPDATA%\Claude\
  1. 1Get the exact server name, package, and API key from your vendor's MCP docs.
  2. 2Open Claude Desktop → Settings → Developer → Edit Config and paste the JSON.
  3. 3Save, fully quit Claude, and reopen — the 🛠 tools icon should appear in the input bar.
06 / Combining servers

The real magic is multiple MCPs in one prompt.

A single MCP makes your model smarter. Three MCPs in one prompt is when it starts to feel like an actual coworker. The model decomposes a request, decides which servers to call in which order, and stitches the results into a single answer — all inside one chat.

The example below is one we run every day at Masset: a sales rep asks for a case study to be sent to a prospect, and three MCPs (Masset, HubSpot, Gmail) cooperate to find the asset, log the engagement, and draft the email.

01 / orchestration
One question, three tools.
Your rep asks one question.
02 / what's happening
00“Find the Memorial Health case study, log it on Sarah's deal, and email it to her.”

You don't have to teach the AI which tool to use

The AI reads each MCP's description, picks the right tool, and decides the order — exactly like a new hire reading their onboarding doc. Your job is to install the right MCPs and let the AI route the work. The more good MCPs you have, the more your team can do without leaving one window.

07 / Vendor strategy

What if my favorite tool doesn't have MCP yet?

Early 2026 was the inflection point — most modern B2B SaaS companies are scrambling to ship MCP support. If a tool you depend on doesn't have it yet, you have three good options.

01 / option
Ask

Email your account manager and reference MCP by name. Vendors are tracking these requests carefully right now — a single nudge from a real customer routinely jumps the roadmap.

02 / option
Wait

Most major B2B tools will have native MCP support by end of 2026. If the tool isn't blocking your team today, this is often the cleanest path.

03 / option
Bridge it

Pipedream and Zapier offer MCP “bridges” for hundreds of apps without native support. Slightly less polished, but works in a pinch.

Five questions to ask vendors who claim MCP support

Once a vendor says yes, here's how to make sure the support is real. Forward this list to whoever runs procurement.

Is MCP available on my plan, or only on a higher tier?
Some vendors gate MCP behind enterprise plans. Get pricing in writing before you commit.
How are permissions handled — does the AI inherit my access, or get its own?
The right answer is: the AI inherits the user's permissions, scoped to that user's session. If the AI gets its own super-key, that's a red flag.
Can you show me audit logs of what the AI did on my behalf?
Every action the AI takes should be logged: who, when, which tool, with what arguments. If the vendor can't show you the log, walk away.
Is authentication done with API keys or OAuth?
OAuth is strongly preferred for shared/enterprise installs — easier to revoke, easier to audit, works with SSO.
Do you support both desktop AI tools and ChatGPT-style remote connectors?
Most enterprise vendors should support both modes. If not, ask why — and what their roadmap is.
08 / Best MCPs in 2026

Eight servers worth installing today.

The MCP registry is enormous and growing weekly. Below is the short list we recommend a B2B operator install on day one. Every one of these is battle-tested, well-maintained, and covers a meaningful slice of daily work.

Masset
Content & enablement
official

Search every approved asset by meaning, pull engagement data, and create shareable links — inside any AI conversation your team is already having.

HubSpot
CRM
official

Pull deal context, log activity, and update contact records mid-conversation. The CRM stops being a separate tab.

Salesforce
CRM
official

Read pipeline, summarize accounts, and update opportunities without leaving Claude or ChatGPT.

Gmail
Email
official

Draft, search, and label messages directly. Especially powerful when paired with a CRM and content MCP.

Slack
Messaging
official

Search channels, summarize threads, and post messages — without leaving the AI conversation.

Notion
Docs & wiki
official

Pull docs into the conversation, draft new pages, and update meeting notes after a call.

Asana
Tasks & projects
official

Create tasks, summarize project status, and assign follow-ups based on the conversation.

Calendly
Scheduling
official

Generate booking links, surface availability, and route prospects to the right rep — inline.

09 / Security checklist

Five questions before you roll this out company-wide.

MCP gives your AI the ability to act on your behalf. That's the whole point — and the whole risk. Forward this checklist to your IT and procurement teams before installing anything beyond a personal pilot.

01
Confirm permissions match yours
Your AI should only see what you can already see — never more. Ask each vendor how this is enforced. The answer should mention the user's session, not a shared service account.
02
Insist on OAuth for remote MCPs
OAuth means access can be revoked in a click and your password never leaves your company. Pasting raw API keys into a settings file is a yellow flag for any shared install.
03
Require human approval on writes
When the AI is about to send an email, update a deal, or post to Slack, the AI tool should ask you first. Most do this by default — but verify before rolling out broadly.
04
Demand audit logs
Every AI action should be logged with timestamp, user, tool, and arguments — exactly like any other system that touches your data. If a vendor can't show you a log, that's a no.
05
Treat MCP installs like vendor onboarding
Same security review, same procurement process. SOC 2 reports, pen tests, DPAs — all the things you already require of new vendors. MCP doesn't change the bar; it just makes more tools subject to it.
10 / FAQ

Common questions.

No. The AI tools your team already uses (Claude, ChatGPT) have MCP built in. The only one-time setup is pasting a snippet into a settings file — usually handled by IT, RevOps, or whoever manages your team's tooling. Once it's in, your team gets the new capabilities for free in every conversation.
Zapier connects two tools through a fixed recipe — “when X happens, do Y.” MCP gives an AI live access to your tools so it can decide what to do, on the fly, based on the conversation. Zapier is rails. MCP is a coworker who can read the rails. They're complements, not replacements.
Only as far as the AI tool itself. The MCP server runs on the vendor's existing infrastructure (in Masset's case, alongside the rest of Masset). Most vendors offer the same data-residency and compliance commitments as the rest of their product — but always check before installing, and route enterprise installs through your usual security review.
Most major B2B tools have native MCP support already, or are about to ship it. If yours hasn't, ask your account manager — vendors are tracking these requests carefully and a single nudge from a real customer routinely jumps the roadmap. In the meantime, Pipedream and Zapier offer MCP “bridges” for hundreds of tools without native support.
No. MCP makes your AI tools dramatically better at working with your existing systems — but those systems stay where they are. Your CRM is still your source of truth. MCP is just how the AI reaches into it.
There's a one-time IT effort to install (and to vet new MCPs the same way you'd vet any new vendor). The bigger ongoing risk is permissions: a poorly-configured MCP can give the AI broader access than the user should have. Section 09 above is the procurement checklist that prevents that.
What's next

Plug your content into every AI conversation your team is already having.

The Masset MCP turns your approved content library into a tool every AI client can call — with permissions, analytics, and audit trails included.

We use cookies to improve your experience and analyze site traffic. Privacy Policy