Model Context Protocol (MCP) is an open standard introduced by Anthropic in November 2024 that lets AI agents connect securely to external tools and data sources through a single, standardised interface.

MCP: A recap

Before MCP, connecting AI to each new system required separate engineering work every time. For a PE firm running DealCloud, SharePoint, Outlook, and a data room, that's multiple projects before touching the long list of third-party tools a fund uses. MCP replaces that with a single standard: any system that supports it can connect to any MCP-compatible AI agent without additional development work.

Since its introduction, MCP has been adopted by every major AI provider - Anthropic, OpenAI, Google, and Microsoft - as well as AWS and Azure. In under 18 months, it has gone from a new open standard to the default infrastructure layer for enterprise AI integration.

How it works

Think of MCP as a universal key card for your firm's data environment. Before it existed, each door - your CRM, your data room, your file storage - needed its own key cut specifically for your AI system. MCP issues one key card that works across every door that supports the standard.

What that looks like in practice: an analyst preparing for an IC meeting would previously open DealCloud to pull relationship history, switch to SharePoint to find the prior sector note, open Salesforce for contact history, and check Granola for meeting notes - manually compiling everything before analysis could even begin. With MCP, a single query surfaces all of it simultaneously, drawn live from each source in one step.

MCP is an access layer, not an intelligence layer. It surfaces the data and the model does the reasoning. Connecting that data to a pattern across 50 prior deals, and making that insight available to every analyst on every future deal, requires a purpose-built intelligence layer on top. MCP is the pipe. What flows through it, and what compounds from it, depends on what's built above it.

In Capsa

Through MCP, Capsa connects securely to the tools and systems firms already rely on - from Granola and Stockpulse to Dropbox, Salesforce, and more - without exposing data outside your existing security perimeter. Every workflow - diligence, IC prep, portfolio monitoring - runs on live, connected data with every insight linked back to its original source for auditability.

What's coming next: ACP

MCP solves the connection between an AI agent and your data. Agent Communication Protocol (ACP) is the emerging answer to the next problem: connections between agents themselves.

If MCP is the key card that gets you into every room, ACP is the intercom that lets everyone inside talk to each other. In a deal context, that means a sourcing agent identifying targets, passing findings to a diligence agent screening them against investment criteria, which hands off to an IC prep agent synthesising the output - each built for its specific task, none requiring a human to stitch outputs together.

ACP is earlier-stage than MCP but follows the same logic: standardise the connection layer, and the workflows that become possible expand significantly. As these standards mature, the deal teams already operating on connected AI infrastructure will be the first to benefit.

Don't miss future insights. Subscribe to the Deal Intelligence Brief here.