What is MCP, and why it's the missing layer between AI and your CRM

#ai #claude #mcp #devops

Last week I made a claim: your AI assistant can't actually read your pipeline.

A lot of people agreed. A few pushed back: "Can't you just use the API?" or "What about RAG?" Fair questions. Let me answer them properly.

Three ways AI can access external data

Approach	Best for	Reality for GTM teams
Direct API calls	Custom integrations	Requires engineering, breaks on schema changes, returns raw data the AI struggles to interpret strategically
RAG	Knowledge bases and SOPs	Great for documents — poor for live structured records like pipeline stages or customer segments
MCP	Live business intelligence	AI calls purpose-built tools in plain language; tools run logic against live data; returns intelligence, not raw rows

MCP isn't better because it's newer. It's better because it was designed specifically for this problem.

What MCP actually is - no jargon

Model Context Protocol is an open standard published by Anthropic. It defines how an AI assistant can discover, call, and receive results from structured tools.

Your AI assistant is an extremely capable analyst. MCP is the secure, structured desk where all the right files are waiting — organised, pre-processed, and ready to reason over.

You ask in plain English. The tool runs the logic. The AI interprets the result.

How it works under the hood (simplified)

Here's the flow when you ask Claude: "Analyze my pipeline health"

Claude receives your message and detects it matches a tool call
It sends a structured request to the Pipeline Health Scoring tool
The tool queries your CRM data, runs velocity and conversion analysis, returns a scored JSON payload
Claude interprets the score, explains what it means, and recommends the next action

No copy-pasting. No screenshots. No engineering middleware.

Artefact MCP: a real-world implementation built for GTM

I spent months translating 15 years of revenue consulting methodology into MCP tools. The result is the Artefact MCP Server — an open-source package that turns Claude into a GTM intelligence advisor.

It ships with 7 tools, works with sample data out of the box (no API key required), and installs in 3 steps:

# Step 1 — Install
pip install artefact-mcp

# Step 2 — Configure
claude mcp add artefact-mcp

# Step 3 — Ask Claude in plain language
"Analyze my pipeline health"
"Who are my ideal customers?"
"What is my dominant growth constraint?"

Artefact MCP vs HubSpot's official MCP server

HubSpot shipped their own official MCP server. It's excellent for CRM read/write access. Artefact MCP brings analytical methodology — ICP triangulation, RFM segmentation, signal detection — that HubSpot's server doesn't include. They're complementary. Run both.

Capability	Artefact MCP	HubSpot MCP
ICP Triangulation	Yes	No
RFM Customer Scoring	Yes	No
Pipeline Health Score	Yes	Partial
Signal Detection	Yes	No
Constraint Analysis	Yes	No
GTM Commit Drafting	Yes	No
CRM Read / Write Ops	No	Yes
Works Without API Key	Yes	No
Methodology Built-in	Yes	No