The open standard that lets AI agents securely access your firm's data — and why it changes everything.

Model Context Protocol (MCP) is an open standard created by Anthropic that defines how AI applications connect to external data sources and tools. Think of it as USB-C for AI: a universal connector that lets any AI agent access any data source through a standardized interface, without custom integration work for every combination.

What is Model Context Protocol?

Model Context Protocol — MCP — is an open standard that Anthropic released to solve a fundamental problem with AI assistants: they are intelligent, but they are blind. Out of the box, a large language model like Claude knows a tremendous amount about the world, but it knows nothing about your firm. It cannot see your client data, your portfolio positions, your advisor pipelines, or your operational metrics.

MCP defines the interface layer that connects AI agents to external systems. It specifies how an AI application sends requests to a data source, how that data source authenticates the request, how results are formatted and returned, and how access is governed and logged. The result is a standardized handshake between AI intelligence and firm-specific data.

Before MCP, connecting an AI tool to a data source required building a one-off integration — custom code specific to that AI tool, that data source, and that use case. Multiply that across every combination of AI tool and data system your firm might want to use, and the engineering burden becomes prohibitive. MCP replaces that matrix of custom integrations with a single, reusable standard.

Open Standard, Not Proprietary Lock-In

MCP is published as an open specification. Any AI provider can build MCP support into their product. Any data platform can expose an MCP server. Because the protocol is open, firms that adopt MCP-compatible infrastructure today are not locked into a single AI vendor — they can switch or expand to any MCP-compatible agent in the future.

Claude was the first major AI model to ship native MCP support, and it remains the most capable MCP implementation available. But the protocol was designed for the industry, not for Anthropic's competitive advantage.

Why MCP Matters for Wealth Management

The Current State: AI Without Data

Today, most advisory firms using AI tools face a fundamental limitation: the AI has no access to the firm's actual data. Advisors wanting to use ChatGPT or Claude to answer client questions, generate reports, or analyze their book of business are forced to copy-paste from their CRM, export spreadsheets from their portfolio system, or type numbers in manually before they can get a useful response.

This workflow creates three serious problems. First, it is slow and error-prone — manual data entry defeats the productivity benefit of AI. Second, it creates compliance risk — copying sensitive client data into a third-party chat interface may violate your data handling policies and regulatory obligations. Third, the AI is only as good as the data it receives, and manually assembled data is always incomplete, always stale, and always a function of what the advisor remembered to include.

Without MCP

Advisor copies data from CRM into ChatGPT

AI answers based on incomplete, manually assembled data

Client data enters a third-party system without governance

No audit trail of what data was shared or with whom

With MCP

Advisor asks Claude a question in plain language

AI queries live firm data through the secure MCP layer

Data stays within your infrastructure; AI receives only results

Every query logged with user, timestamp, and scope

The Opportunity: AI That Knows Your Firm

MCP transforms AI from a general-purpose writing assistant into a firm-specific intelligence layer. When Claude can query your actual data, the quality of its outputs changes fundamentally. Instead of generic advice based on your description of a client situation, it can analyze the actual portfolio. Instead of boilerplate review language, it can generate a summary grounded in the real account history. Instead of estimated figures, it can return exact numbers from your warehouse.

This is the competitive asymmetry that MCP creates: firms whose AI has access to firm data will outperform firms whose AI does not. The gap compounds over time as AI capabilities improve and the value of clean, accessible data increases.

How MCP Works (Simplified)

The MCP architecture has three components: the AI agent (Claude), the MCP server (a secure middleware layer that speaks both AI and data), and your data sources (your Snowflake warehouse, CRMs, portfolio systems). Here is what happens when an advisor asks a question.

1

The advisor asks a question "What is our total AUM broken down by custodian this quarter?"

2

Claude recognizes a data need The AI determines it needs to query firm data to answer accurately, and sends a structured request to the MCP server.

3

The MCP server authenticates and routes The MCP layer verifies the user's permissions, translates the request into a data query, and executes it against the appropriate data source — in Milemarker's case, your Snowflake warehouse.

4

Results return to Claude The query results — structured data scoped to exactly what was asked — are returned to the AI agent. Raw records never leave your infrastructure.

5

Claude formats the answer The AI synthesizes the query results with its language capabilities and returns a clear, formatted answer to the advisor — complete with numbers, breakdowns, and context.

6

The interaction is logged Every MCP query is recorded: who asked, what was queried, when, and what data was accessed. The audit trail is available for compliance review.

The entire exchange happens in seconds. From the advisor's perspective, they asked a question and got an accurate answer. The MCP layer handles authentication, permissioning, query translation, and audit logging invisibly.

MCP vs. Traditional API Integrations

Wealth management firms have long used APIs to connect systems — custodian feeds, portfolio data, CRM sync. So it is reasonable to ask: what does MCP add that a traditional API does not provide?

The difference is direction and standardization. Traditional APIs are built point-to-point: a Salesforce integration delivers Salesforce data to a specific application using Salesforce's API specification. If you want a different application to access Salesforce, you build another integration against the same API. If you want Salesforce data in a new AI tool, you build yet another integration — this time tailored to how that AI tool expects to receive data.

MCP inverts this. Instead of building an integration for every AI-plus-data combination, you build one MCP server per data source. Any MCP-compatible AI agent can then use any MCP-compatible data source without additional engineering. The protocol standardizes the interface so the combination matrix collapses from N-times-M integrations to N-plus-M.

Integration effort

• Traditional API Integration: Custom code for every AI + data source combination

• MCP: One MCP server per data source; works with any MCP-compatible AI

AI tool flexibility

• Traditional API Integration: Locked to the AI tool the integration was built for

• MCP: Any MCP-compatible AI agent can connect immediately

Governance model

• Traditional API Integration: Depends on implementation; often limited

• MCP: Auth, permissioning, and audit logging built into the protocol

Maintenance burden

• Traditional API Integration: Each integration maintained separately; breaks when either side changes

• MCP: Protocol changes managed centrally; data source changes update once

Time to new AI tool

• Traditional API Integration: Weeks to months of integration work per tool

• MCP: Hours; the interface is already standardized

Data exposure model

• Traditional API Integration: Varies; often raw data transfer

• MCP: AI receives query results, not raw data access

This is the shift from bespoke to standardized — the same transition that transformed software development when REST APIs replaced custom SOAP implementations, or when USB replaced proprietary connectors. MCP is that standardization layer for AI-to-data connectivity.

Security and Compliance

For registered investment advisors and broker-dealers, the question of AI data access is inseparable from the question of regulatory compliance. Any system that touches client data must satisfy GLBA privacy requirements, SEC cybersecurity rules, and FINRA's data governance expectations. MCP was designed with these constraints in mind.

Data Never Leaves Your Infrastructure

The most important security property of MCP is architectural: the AI agent never has direct access to your database. The MCP server sits between the AI and your data, executing queries and returning results. Your client records, portfolio data, and financial information stay in your Snowflake warehouse. The AI receives only the answer to the specific question it asked — not open-ended access to browse records.

Role-Based Access Control

MCP enforces permissioning at the protocol level. An advisor authenticated as a wealth manager can query data about their own clients, but cannot access another advisor's book of business without explicit permission. Operations staff can pull firm-wide aggregates without seeing individual account details. Administrators define access scopes; MCP enforces them on every query.

Full Audit Trails

Every MCP query is logged: the requesting user, the question asked, the data scope queried, the timestamp, and the result returned. For firms subject to regulatory examination, this creates a complete record of AI-assisted data access — comparable to the audit logs you already maintain for direct database queries and system logins. Compliance teams can review the log, verify appropriate use, and produce evidence of governance for regulators.

Authentication at Every Layer

MCP requires authentication before any query executes. Users authenticate to Claude through your identity provider. Claude authenticates to the MCP server using a server-issued credential. The MCP server authenticates to your Snowflake warehouse using database credentials that never leave your infrastructure. No anonymous queries. No credential sharing. No path for unauthorized access that bypasses the permissioning layer.

What You Can Do With MCP in Wealth Management

The practical value of MCP becomes clear when you see the questions it makes answerable. Here are examples of what advisors and operations teams can ask Claude when it has MCP access to firm data.

"Pull a report on all clients with more than $1M in fixed income exposure."

"Show me advisor productivity this quarter — AUM, revenue per advisor, and client acquisition."

"Draft a quarterly review summary for the Johnson household based on their actual account data."

"What is our total AUM broken down by custodian and asset class?"

"Which clients haven't had a review meeting in over 12 months and have accounts over $500K?"

"Identify clients who may be candidates for tax-loss harvesting based on current positions and cost basis."

"Summarize new business activity from last month — new accounts opened, AUM added, and referral sources."

These are not hypothetical future capabilities. They are available today through Milemarker's production MCP integration. The queries run against live Snowflake data. The answers come back in seconds. No dashboard navigation. No export and reimport. No manual data assembly. The advisor asks a question and gets an accurate answer grounded in real firm data.

Milemarker's MCP Implementation

Milemarker is one of the first WealthTech platforms to ship a production MCP integration. While most wealth management technology vendors are still evaluating MCP conceptually, Milemarker has been running it in live production environments — connecting Claude to client data warehouses and returning real query results to advisors and operations teams.

Built on Snowflake

Milemarker's data platform is built on Snowflake, the cloud data warehouse that has become the industry standard for wealth management analytics. All 130+ integrations — CRMs, custodians, portfolio management systems, planning tools, compliance platforms — flow into Snowflake as a single, normalized data warehouse. Milemarker's MCP server sits on top of that warehouse, making the entire consolidated dataset available to Claude through a governed, authenticated interface.

This architecture matters because it means Claude is not querying fragmented, disconnected data sources. It is querying a single, clean, reconciled warehouse where your client data, portfolio data, and operational data are already joined and normalized. The quality of the AI's answers is a direct function of the quality of the underlying data — and Milemarker's unified warehouse provides the highest-quality data foundation available in WealthTech.

130+ Integrations, Day One

Building an MCP integration from scratch requires first solving the data consolidation problem: getting clean, normalized data from dozens of disparate sources into a queryable format. Milemarker's 130+ pre-built connectors solve that problem before MCP is even introduced. Firms that implement Milemarker's data platform automatically get MCP-ready data — they do not need to build the warehouse before they can use the AI.

Governed for Regulated Environments

Milemarker's MCP implementation was designed for the regulatory requirements of registered investment advisors and broker-dealers. Role-based access controls are configurable at the advisor, team, and firm level. Audit logs capture every query in a format compatible with regulatory examination. Data never leaves your Snowflake warehouse. The implementation satisfies the compliance requirements that make AI adoption possible for regulated firms.

For full technical details on Milemarker's MCP integration — including setup, available tools, and example queries — see the MCP integration page.

The Future: MCP as the Industry Standard

MCP is early. The specification was published less than a year ago. The number of production MCP integrations in wealth management can be counted on one hand. But the trajectory is clear — and it mirrors what happened with every prior data standardization effort in the industry.

When custodians adopted standardized data feeds, firms that integrated early had years of analytical advantage before the rest of the industry caught up. When portfolio accounting systems adopted common data schemas, firms that moved first built the reporting infrastructure that slower competitors are still catching up to. MCP is that moment for AI data access.

What Adoption Looks Like

As more platforms adopt MCP — CRMs, portfolio systems, custodian portals, planning tools — advisory firms will gain the ability to use any MCP-compatible AI agent with any MCP-compatible data source. The integration work done today compounds in value as new AI capabilities emerge. A firm that has its data MCP-ready today will be able to deploy the next generation of AI tools without another integration project.

01

More AI Agents

As additional AI providers add MCP support, firms with MCP-ready infrastructure gain access to new tools without new integrations.

02

More Data Sources

As WealthTech vendors ship MCP servers, the range of firm data accessible to AI expands without additional engineering work.

03

Richer Capabilities

MCP-connected AI will gain the ability to take actions — not just answer questions — as the protocol matures toward agentic use cases.

04

Regulatory Clarity

As MCP adoption grows, regulators will develop clearer guidance for AI-assisted data access — likely favoring the governed model MCP provides.

First-Mover Advantage Is Real

Advisory firms that implement MCP-connected data infrastructure today are not just buying a feature — they are building the operational muscle of AI-native wealth management. The advisors and operations teams who learn to query their data through Claude today will be dramatically more effective than those who start that learning curve two years from now. The institutional knowledge of how to use AI with firm data is itself a competitive asset.

Milemarker's position as one of the first production MCP implementations in WealthTech means that the firms on its platform today are accumulating that advantage while the rest of the industry catches up.

Frequently Asked Questions

What is Model Context Protocol (MCP)?

Model Context Protocol (MCP) is an open standard created by Anthropic that defines how AI applications connect to external data sources and tools. It gives AI agents like Claude a standardized, secure interface to query databases, call APIs, and retrieve real-time information from external systems — without requiring custom integration code for every AI-plus-data-source combination. Think of it as USB-C for AI: a universal connector that works across providers and platforms.

Is MCP secure for regulated financial firms?

Yes. MCP is designed with security as a first-class concern. The protocol includes built-in authentication, role-based permissioning, and audit logging. The AI agent receives answers derived from your data — not direct access to raw records. Your data remains within your own infrastructure. For firms subject to FINRA, SEC, and GLBA regulations, MCP's governance model provides the controlled, auditable access that compliance requires.

Which AI tools support MCP?

Claude (Anthropic) has native MCP support and is the leading MCP-compatible AI today. The MCP specification is open source, so other AI providers are actively adding support. Because MCP is a standardized protocol rather than a proprietary integration, any MCP-compatible AI agent can connect to any MCP-compatible data source — the same way any USB-C device works with any USB-C cable.

Do I need Milemarker to use MCP?

No. MCP is an open protocol anyone can implement. However, building and maintaining an MCP server that connects to 130+ wealth management data sources — CRMs, custodians, portfolio systems, planning tools — requires significant engineering investment. Milemarker's production MCP integration gives advisory firms that capability without the build cost, and connects to the Snowflake data warehouse where your firm's data already lives.

How is MCP different from a traditional API integration?

Traditional API integrations require custom code for every combination of AI tool and data source. If you want ChatGPT to query Salesforce, you build a Salesforce integration. If you want Claude to query Orion, you build an Orion integration. With MCP, the interface is standardized: build one MCP server per data source, and any MCP-compatible AI agent can use it immediately. This eliminates the exponential engineering work of N-times-M bespoke integrations and replaces it with N-plus-M standardized ones.

How long does MCP implementation take?

Building an MCP server from scratch for a complex data environment can take months. With Milemarker's pre-built MCP integration — which already connects to 130+ wealth management data sources via Snowflake — firms can be operational in weeks rather than months. The implementation timeline depends primarily on data availability in your Snowflake warehouse, not on integration engineering. Firms already on the Milemarker platform can typically activate MCP access in days.

What can I actually do with MCP in wealth management today?

With Milemarker's MCP integration, advisors and operations teams can ask Claude questions like: "Show me all clients with more than $1M in fixed income", "What is our total AUM broken down by custodian?", "Which advisors are below their productivity targets this quarter?", or "Draft a quarterly review summary for the Johnson household." The AI agent queries your live Snowflake data, formats the answer, and returns it in plain language — no dashboards, no exports, no waiting.

Does the AI see my clients' raw data?

The AI agent receives structured query results scoped to the specific question asked, subject to the role-based access controls you configure. It does not have open-ended access to browse raw records. Milemarker's MCP implementation enforces permissioning at the data layer, so an advisor querying their own book of business cannot inadvertently access another advisor's client data. Every query is logged with the user identity, question asked, and data scope returned.

Related guides

Part of the AI for Wealth series:

• Claude for Financial Advisors: AI That Knows Your Clients, Portfolio, and Practice

• Claude for Financial Services: How Wealth Management Firms Use AI With Real Client Data

• AI Data Security for Wealth Management: How to Use Claude Safely With Client Data

• AI Copilot for Wealth Management

• Claude for RIA Operations: AI-Powered Answers for COOs, Ops Managers & Firm Leadership

• AI for Wealth Management Starts with Data, Not Agents