Model Context Protocol: Why This Boring Infrastructure Matters

At every firm I advise, the bottleneck isn't capability—it's integration. Your AI tools live in silos. Claude doesn't talk to your CRM. ChatGPT doesn't see your documents. Slack doesn't pass context to your analysis models.

By July 2025, Model Context Protocol (MCP) is starting to solve this. It's unglamorous infrastructure, but it's the foundation that makes enterprise AI actually work.

What Is MCP?

Model Context Protocol is an open standard for connecting LLMs to data sources, tools, and external systems. It's like an adapter that lets AI models understand and access your firm's information ecosystem.

Think of it like this: Instead of pasting information into Claude manually, MCP lets Claude directly access:

• Your CRM (Salesforce, HubSpot, Pipedrive)
• Your document systems (SharePoint, Notion, Google Workspace)
• Your databases and data warehouses
• Your project and time tracking tools
• Your communication platforms (Slack, Teams)
• Custom internal systems

Without MCP, you have to manually copy data into the AI. With MCP, the AI has native access to live, current information.

Why This Matters for Professional Services

Three specific ways MCP changes the game:

1. Context That's Actually Current

Most AI usage in firms relies on stale context. You copy a client file from three weeks ago, paste it into Claude, ask for analysis. Now the client has changed strategy and you're analyzing outdated information.

With MCP, when you ask Claude to analyze a client, it directly accesses the current CRM record, recent emails, project status, and financials. The context is live.

This is especially valuable for advisory work where decisions depend on current information.

2. Agents That Can Actually Do Things

Earlier I wrote about AI agents and their limitations. MCP is what unlocks more capable agents. An agent with MCP access can:

• Read a client request from email
• Pull relevant information from the CRM
• Check project status and resource availability
• Create a proposal or timeline
• Route it to the right stakeholder

Without MCP, the agent would need custom integrations to each of those systems. With MCP, it's a standard connection.

3. Reduced Manual Integration Work

Every firm I work with has spent thousands building custom integrations between tools. Zapier connections that don't quite work. Custom APIs that need maintenance. MCP standardizes this.

Less custom code means faster deployment, lower maintenance, and less technical debt.

The Practical State (July 2025)

By July 2025, MCP is emerging but not yet ubiquitous:

What Exists

• Claude and other Anthropic tools have native MCP support.
• OpenAI is implementing MCP compatibility.
• Server implementations exist for major systems: Salesforce, Google Workspace, Slack, and others.
• Open-source community is building more integrations.

What's Immature

• Broader vendor adoption is still happening. Not every SaaS tool has an MCP server yet.
• Custom integration is still required for some internal systems.
• Standards for security and authentication are evolving.

What's Coming

By end of 2025, I expect:

• Broader adoption. Most major business tools will have MCP servers.
• Standardized security. Clear protocols for authentication and authorization.
• Platform support. Google, Microsoft, and others will build MCP into their platforms.

How This Changes Your AI Stack

If you're building an AI capability now (July 2025), MCP should influence your choices:

For Foundation Models

Claude has the best MCP support. If MCP integration is important to your stack, that's a point in Claude's favor. OpenAI is catching up, but Claude is ahead.

For Integration

If you're choosing between custom API integration and MCP, choose MCP whenever it exists. It's faster, more maintainable, and more standardized.

For Tool Selection

When evaluating SaaS tools, ask: "Does this tool have an MCP server?" It's increasingly important for tools you want to integrate with AI.

The Real Advantage

Here's what MCP enables that's actually valuable for professional services:

Research and Advisory Work

When writing a strategy recommendation or competitive analysis, the consultant can ask Claude: "Analyze this client's situation against competitors. Pull the CRM context, recent financials, and strategic plans. What are the key issues?"

Claude directly accesses the information instead of the consultant hunting for it manually.

Project Management

A project manager can ask: "Who's available for the new engagement? What's our current capacity? When can we start?" The AI checks project systems directly, not from memory.

Business Development

A business development lead can ask: "Tell me about this prospect. What's their history with us? Recent interactions? Who's the best person to lead the pitch?" The AI pulls CRM data in real-time instead of from notes.

Why Infrastructure Matters

Most firms focus on the exciting AI part (models, capabilities, use cases). But infrastructure is what makes it work at scale. MCP is unsexy infrastructure that solves a real problem.

By July 2025, the firms getting the most value from AI are the ones investing in integration and data infrastructure, not just model capability.

What to Do Now

If you're deploying AI in mid-2025:

1. Check if MCP servers exist for your critical systems (Salesforce, Slack, etc.). If yes, plan to use them.
2. For systems without MCP, start building standard API integrations with AI access in mind.
3. Choose AI platforms (Claude) that have strong MCP support.
4. Build governance around who can access what through MCP. This is more powerful than API access and needs clear controls.

By 2026, MCP will be table stakes for enterprise AI. Getting ahead of it now is a small competitive advantage.