The integration problem
Right now, your AI exists in a box. You open a chat window, paste in some text, get a response, copy it back out. The AI cannot see your CRM, your document management system, your email, or your databases.
This is like having a brilliant analyst who sits in a separate building with no access to your systems. Every time they need information, someone has to print it out and carry it across the street. Every time they produce something, someone has to manually enter it back into your systems.
Model Context Protocol (MCP) solves this. It is the standard that lets AI connect directly to the tools where your work already lives.
The USB-C analogy
Before USB-C, every device had its own proprietary charger. Your phone, laptop, tablet, camera — all different cables. USB-C replaced them all with one universal standard.
MCP does the same thing for AI integrations. Before MCP, connecting AI to each tool (Salesforce, Bloomberg, your database, Slack) required a custom integration. Each one was expensive to build and maintain.
With MCP, there is one standard protocol. Any AI that speaks MCP can connect to any tool that speaks MCP. Build the connector once, and it works everywhere.
| Without MCP | With MCP |
|---|---|
| Copy data from Bloomberg, paste into AI chat | AI reads directly from your data feeds |
| AI produces analysis, you manually update the CRM | AI updates the CRM record as part of the workflow |
| Search your email for context, paste relevant threads | AI searches your email and incorporates relevant context automatically |
| Export AI output, format it, upload to document management | AI saves directly to your document management system |
The shift is from AI as a separate tool to AI as an integrated layer that works across all your existing systems.
Why is MCP compared to USB-C?
The business case for MCP
Cost reduction: Every custom AI integration your firm builds costs money to develop and maintain. MCP reduces this to a standard, reusable pattern. One connector per tool, usable by any AI platform that supports MCP.
Speed to value: Without MCP, connecting AI to a new data source might take weeks of engineering. With MCP, pre-built connectors for common tools (Salesforce, Google Workspace, Slack, databases) are available and can be deployed in days.
Vendor flexibility: MCP is an open standard, backed by Anthropic, and adopted by OpenAI, Google, Microsoft, and Salesforce. This means you are not locked into a single AI vendor. If you build MCP connectors for your systems, they work with any AI platform that supports the protocol.
The market signal: The MCP ecosystem reached $1.8 billion in 2025. Over 1,000 pre-built connectors are available. This is not an experimental technology — it is becoming the standard integration layer for enterprise AI.
MCP in financial services
Here is what MCP makes possible for your firm — not in theory, but in practice.
Cross-system deal workflows: Without MCP, you pull portfolio data from your database, paste it into Claude for analysis, copy the results into a memo, email it to the team, and update the CRM manually. With MCP, you ask Claude to "pull the latest financials for Portfolio Company X, compare to the 100-day plan, draft a quarterly update, and save it to the deal room." Claude connects to your database, does the analysis, produces the document, and saves it — in one workflow.
Research automation: Without MCP, you manually check news feeds, download reports, and synthesise information. With MCP, Claude monitors your configured news sources, pulls relevant articles, cross-references with your portfolio, and surfaces a daily briefing — automatically.
Compliance monitoring: Without MCP, your compliance team manually reviews regulatory updates and assesses impact. With MCP, Claude connects to regulatory feeds, identifies changes relevant to your firm's activities, and drafts impact assessments — flagging items that need human review.
What is the common pattern across all these MCP use cases?
What leaders need to know
You will not build MCP integrations yourself. Your technology team or an implementation partner will handle the technical work. But as a business leader, you need to be able to do three things.
1. Identify high-value integration points. Ask: "Where does my team spend time moving information between systems?" Those manual hand-offs are the integration points where MCP delivers the most value.
| System | Integration Value |
|---|---|
| CRM (Salesforce, DealCloud) | Auto-update deal records, pull client context into AI conversations |
| Document management (iManage, SharePoint) | Save AI outputs directly, search and retrieve documents |
| Data providers (FactSet, Bloomberg, PitchBook) | Pull financial data directly into analysis |
| Email (Outlook, Gmail) | Search email for deal context, draft and send communications |
| Messaging (Slack, Teams) | Surface AI analysis in team channels, trigger workflows |
| Databases (internal) | Query portfolio data, update records, run reports |
2. Evaluate implementation partners. When your technology team proposes an MCP implementation, ask:
- "Which systems are we connecting first, and why those?"
- "What is the expected time savings per workflow?"
- "How are we handling authentication and access controls?"
- "What is the testing plan before we go live?"
- "What happens if the MCP connection goes down — is there a fallback?"
Security and governance
3. Govern the integrations. MCP integrations should be subject to the same governance as any system integration:
- Access controls: Which users can access which connectors?
- Audit trails: All AI actions through MCP should be logged
- Data classification: Does the connector handle MNPI? Client PII? Regulated data?
- Approval workflow: Who authorises new MCP connections?
MCP's design includes a permission model built for enterprise use:
- Authentication: Each connector authenticates separately. Your AI does not get blanket access to everything.
- Scoped permissions: You define exactly what the AI can do — read-only access to some systems, read-write to others.
- Audit logging: Every action the AI takes through MCP is logged and attributable.
- User-level controls: Different team members can have different MCP access levels.
This is not a "give AI the keys to everything" approach. It is a controlled, governed integration layer.
Getting started
If you are interested in MCP for your firm, here is the practical path:
- Identify 2-3 high-value workflows where manual data movement is the bottleneck
- Check connector availability — many common tools already have MCP connectors
- Run a pilot with a small team on a specific workflow
- Measure the impact — time saved, errors reduced, throughput increased
- Scale based on results — expand to additional workflows and teams
The firms that are moving fastest are not trying to connect everything at once. They are picking one high-impact workflow, proving the value, and expanding from there.
Key takeaways
- MCP is USB-C for AI — one standard protocol that connects AI to all your tools.
- The value is in eliminating hand-offs — every time your team manually moves data between systems, that is an MCP opportunity.
- Pre-built connectors exist for most common enterprise tools. You are not starting from scratch.
- Governance applies — MCP integrations need the same access controls, audit trails, and approval workflows as any system integration.
- Start with one workflow — prove the value, then scale.
You understand the tools. Now it is time to make the case to your leadership. The next module covers building the business case for AI adoption.
Module 7 — Knowledge Check
What problem does MCP solve?
Why is MCP compared to USB-C?
As a business leader, what is the most important question to ask about MCP implementation?
