Skip to content
Primores AI wiki

Model Context Protocol (MCP) — Connecting AI to Your Systems

Model Context Protocol (MCP)

TL;DR: MCP is the “USB-C for AI” — an open standard that lets Claude (and other AI systems) connect to your databases, tools, and internal systems. Instead of copy-pasting data into prompts, MCP gives AI direct, secure access to your information sources.

What Is MCP?

MCP (Model Context Protocol) is an open-source standard for connecting AI applications to external systems:

  • Data sources: Local files, databases, CRM, ERP
  • Tools: Search engines, calculators, APIs
  • Workflows: Specialized prompts, automation sequences

Think of it like USB-C provides standardized hardware connections — MCP provides standardized AI connections.

Why It Matters

The Problem Without MCP

  • Copy-paste data into prompts manually
  • Re-explain context every conversation
  • Can’t access real-time business data
  • Each integration requires custom development

With MCP

  • AI accesses live data directly
  • Context persists across sessions
  • Build once, integrate everywhere
  • Secure, permissioned access

Core Concepts

MCP has three primitives:

PrimitiveControlled ByPurpose
ToolsModelActions the AI can take
ResourcesApplicationData the AI can access
PromptsUserSpecialized instructions

What MCP Enables

Personal AI assistants that access your calendar, notes, and files for personalized responses.

Development tools where Claude Code generates apps using Figma designs directly.

Enterprise chatbots connecting multiple databases for natural language data analysis.

Automation where AI executes end-to-end workflows across systems.

Ecosystem Support

MCP is supported across major AI tools:

  • AI Assistants: Claude, ChatGPT
  • Development: VS Code, Cursor, Claude Code
  • Platforms: Many third-party integrations

Build an MCP integration once → works everywhere.

Enterprise Integrations

HubSpot Connector

Native integration allowing Claude to:

  • Access, analyze, create, update CRM records
  • Log activities, tasks, notes directly
  • Respect HubSpot user permissions automatically

Use case: Generate tailored pitches using actual prospect data, analyze call transcripts, create follow-up tasks.

Salesforce Integration

Via Claude Cowork and MCP:

  • Read customer data for personalization
  • Create records from conversation context
  • Analyze pipeline automatically

Xero Partnership

Multi-year partnership bringing Claude into accounting:

  • Track cash flow automatically
  • Flag unpaid invoices
  • Analyze revenue/profit performance
  • Suggest financial actions

Inside Xero: AI assistant JAX powered by Claude reasoning Inside Claude: Access Xero financial data for analysis and planning

Notion Integration

Connect Claude to Notion databases:

  • Pull strategy data for presentations
  • Access meeting notes and best practices
  • Build documents from stored knowledge

Implementation Approaches

Desktop Extensions (Easiest)

  1. Open Claude Desktop → Settings → Extensions
  2. Browse Anthropic-reviewed extensions
  3. Click to install — no JSON configuration needed

MCP Servers (Technical)

For custom integrations:

  • Build server exposing your data/tools
  • Follow MCP specification
  • Deploy alongside your infrastructure

Resources:

Security Model

Data Privacy

  • Financial data shared only for specific session
  • Proprietary data never used for model training
  • Respects source system permissions (e.g., HubSpot user roles)

Enterprise Controls

  • Permission boundaries configurable
  • Audit trails available
  • Compliance-ready architecture

Real-World Example: Brainlabs

Setup: MCP connection from Claude Cowork to Notion

Skill created: Presentation generator

Workflow:

  1. Employee asks Claude for client presentation
  2. Claude queries Notion via MCP for client data, meeting notes, best practices
  3. Claude assembles slide deck following agency templates
  4. Employee reviews and sends

Before: Hours of manual assembly After: Minutes with AI assistance

Getting Started

For Non-Technical Users

  1. Use pre-built connectors in Claude Desktop
  2. Extensions require no coding
  3. Start with productivity tools (Calendar, Notion, Slack)

For Technical Teams

  1. Review MCP documentation
  2. Take Anthropic Academy MCP courses
  3. Build custom servers for internal systems
  4. Deploy with appropriate security controls

MCP vs. Traditional Integrations

TraditionalMCP
Custom per AI toolBuild once, works everywhere
Point-to-pointStandardized protocol
ProprietaryOpen source
Static dataReal-time access

MCP + Skills: The Complete Solution

MCP provides connectivity — but connectivity alone isn’t enough. Users need workflow guidance.

The Kitchen Analogy

LayerRoleExample
MCPThe professional kitchenAccess to Notion, Linear, Salesforce tools
SkillsThe recipesHow to use those tools effectively

Without Skills

  • Users connect MCP but don’t know what to do next
  • Support tickets asking “how do I do X with your integration?”
  • Each conversation starts from scratch
  • Inconsistent results because users prompt differently

With Skills

  • Pre-built workflows activate automatically when needed
  • Consistent, reliable tool usage
  • Best practices embedded in every interaction
  • Lower learning curve for your integration

Building MCP-Enhanced Skills

Skills can coordinate multiple MCP calls in sequence:

### Phase 1: Design Export (Figma MCP)
1. Export design assets from Figma
2. Generate design specifications


### Phase 2: Task Creation (Linear MCP)
1. Create development tasks
2. Attach asset links to tasks


### Phase 3: Notification (Slack MCP)
1. Post handoff summary to #engineering

Example: Sentry’s sentry-code-review skill automatically analyzes bugs in GitHub PRs using Sentry’s error monitoring data via MCP.

See tools/claude-skills for the complete guide to building skills.

Key Takeaways

  • MCP is the standard for connecting AI to enterprise systems
  • “USB-C for AI” — build once, integrate everywhere
  • Pre-built connectors for HubSpot, Salesforce, Xero, Notion
  • Security built in: permissions, audit trails, no training on your data
  • Enables Level 3-4 automation (AI working with your actual systems)
  • MCP + Skills = tools + recipes = complete solution for users

Sources