Blog

What an MCP Server Actually Does for Insurance Agencies

Learn what an MCP server is, what your agency can ask Claude, ChatGPT, or Gemini once its data is connected, and what to check before you connect one.

Mav Team

Most agency reporting questions die in an export. You want to know how many transfers connected, which quoted leads went quiet, or which lead source is earning its spend. The answer lives in your CRM, your AMS, and your lead tools, so somebody pulls a CSV and builds a spreadsheet.

You can ask ChatGPT, Claude, or Gemini about your agency's lead data instead, as long as the system that holds it offers an MCP server. The server lets the assistant look things up in that system, usually after you sign in with your own login. You ask in plain English and get the answer from the source.

It also raises fair questions about access, security, and compliance. Below: what MCP is, what agencies ask it, what to check first, and how setup works.

Key Takeaways

  • You can ask an AI assistant about your agency's leads, transfers, calls, and conversions when the system that holds that data offers an MCP server, with no export in between.

  • MCP is an open standard, introduced by Anthropic in 2024, that lets AI assistants like Claude, ChatGPT, and Gemini read data from the business systems you already use.

  • Read-only access limits what the AI can change, but disclosure and prompt-injection risks still need review.

  • Ask every vendor about sign-in, account scoping, audit logs, and exactly which data the server exposes.

  • The NAIC AI Model Bulletin is written for insurers, so confirm with your carrier and counsel how it reaches your agency.

What Is an MCP Server?

An MCP server is the software that lets an AI assistant look things up in one of your systems, like your CRM or your lead platform. MCP stands for Model Context Protocol. Anthropic introduced it as an open standard on November 25, 2024, for building two-way connections between data sources and AI tools. It has two sides. An MCP server exposes a system's data, and an MCP client, the AI app, connects to that server.

Think of the server as a translator. Your system speaks its own language, the AI assistant speaks another, and the MCP server sits between them. Any assistant that speaks MCP can then ask that system questions.

In Claude and ChatGPT settings, a server you've added shows up as a connector, and Gemini lists it as a connected app, so you'll see all three names for the same connection.

MCP also works well beyond Claude. On December 9, 2025, Anthropic donated MCP to the Linux Foundation's new Agentic AI Foundation, co-founded with Block and OpenAI. In that same announcement, Anthropic counted more than 10,000 active public MCP servers. It also listed ChatGPT, Gemini, Microsoft Copilot, Cursor, and Visual Studio Code as adopters.

For an agency owner, that means one vendor's server can work with whichever of those assistants your team already uses.

Why Does MCP Matter to an Agency Owner?

MCP gets you reporting without exporting. A sales manager asks how many leads replied last week. An owner asks which producer has the most quoted-not-sold leads sitting open. Each answer comes straight from the system of record, in a sentence.

That changes who gets to ask. Today the person who knows the reporting menus answers the questions. With MCP, anyone with the right access can ask, and the AI does the lookup.

The carriers you're appointed with already use AI heavily. NAIC surveys report current or planned AI use by about 88% of auto insurers, 70% of homeowners insurers, and 58% of life insurers. Those figures come from surveys published between 2022 and 2023 and summarized in NAIC testimony (NAIC, 2025). They describe carriers, and they say nothing about agencies.

Insurance vendors have started shipping MCP servers too. Socotra released one in September 2025 so insurers can connect AI agents to its core insurance platform. Fenris offers an MCP server for its insurance data services through an early access program, compatible with Claude, ChatGPT, and Gemini.

On the agency side, offerings are moving fast. As of November 2026, the reliable way to know whether your AMS or CRM has an MCP server is to ask the vendor directly.

What Can You Ask Once Your Agency Data Is Connected?

The best questions are the ones you already ask in your Monday meetings. Here's what that looks like with Mav's MCP server, which connects your Mav account to Claude, ChatGPT, or Gemini. Mav texts every new lead first, follows up, qualifies, and warm-transfers ready leads to a licensed producer. The MCP connection lets you ask about all of that activity.

Lead and Transfer Performance

Start with volume and handoffs. Ask "How many transfers connected this week, and why didn't the rest?" and the assistant looks it up in your Mav account. Follow with "Which leads talked with Mav but were never transferred?" to find the conversations a producer should check on.

If your team tracks transfers and callbacks closely, this is the fastest way to see the week without opening a dashboard.

Call and Conversation Review

Transcripts are where coaching happens. Before a producer calls a lead back, ask "Read me the transcript of the last call with this lead" or "Summarize every text and call with this lead." The producer walks in knowing the vehicles, the price objection, and what the lead already said.

Managers can do the same across a day of calls. Ask "Summarize what leads objected to on yesterday's calls," then talk to producers about the actual words.

Quotes and Sales

Quoted-not-sold is where premium leaks. Mav's connection reads conversions marked Quoted and Sold, so you can ask "Which quoted leads haven't been marked Sold yet?" and see the open quotes that have sat too long.

Lead Source Performance

Every lead source looks good on the invoice. Ask "Rank my lead sources from contacts to sold for the past month, with cost per transfer," and you'll see which vendors turn into quotes and policies, and where to put next month's lead spend. From there, ask for a closer breakdown of any one provider in your marketing sources.

Recaps and Dashboards

Ask for "a morning digest of yesterday vs. the day before" or "a Monday summary I can paste into Slack," and the assistant pulls the numbers and writes it up. You can also ask it to build a dashboard from your Mav data, with rates next to counts and a split by lead source.

Mav calculates every number on its side and hands your AI the result to report. The server can look up contacts, count transfers, read call transcripts, and tally clicks and conversions. It reads leads and party lines on Classic Mav and contacts and phone calls on Mav Next, and answers with both when your account has both.

What Should an Agency Check Before Connecting an MCP Server?

Before you connect one, get clear answers on what the AI can change, which account and data it reaches, how it handles text written by people outside your agency, and how your AI plan and compliance obligations apply.

Read-Only Access Limits What the AI Can Change

Each vendor decides what its server lets the AI do. Ask whether it can only look things up or can also send, edit, or delete.

Mav's connection is read-only. The AI can look up contacts, texts, calls, transcripts, clicks, and conversions. It can't text, call, or change anything in your account. Card and Social Security numbers are redacted in transcripts and texts before they reach your AI.

Read-only reduces the risk of unwanted changes. Data still flows to the assistant, so the questions below still apply.

Sign-In and Account Scoping Decide What the AI Sees

The MCP authorization spec makes authorization optional. When it is supported, the spec builds it on OAuth 2.1. It also says connections should follow the principle of least privilege, requesting only the scopes they need, and that servers must validate tokens issued specifically for them.

In plain terms, a server using MCP isn't automatically locked down. Ask the vendor how you sign in, what access the sign-in grants, and whether it matches the permissions your team already has.

With Mav, you sign in with your Mav login and 2FA, and the connection is tied to the account you pick. For agencies with sub-accounts, the account owner can see every sub-account, and members see only their own data. Mav logs every request and caps usage per user, and you can disconnect anytime from your assistant's connector settings.

Prompt Injection and Tool Poisoning Are the Risks to Know

The OWASP MCP Top 10, still in beta, lists the risks security teams watch. Those include tool poisoning, privilege escalation through scope creep, missing audit logs, and unapproved "shadow" MCP servers. OWASP also flags prompt injection, where the attack payload is plain text the model reads.

That matters in insurance because your data is full of text from strangers. A lead's reply is text the AI reads, so Mav's server tells your AI to treat texts and transcripts from leads as information, never as instructions. A read-only connection also gives a planted instruction no way to text or call anyone, but it can still skew an answer. Check any number that drives a staffing or spend decision against your dashboard.

Your AI Plan Covers What Lands in a Chat

What you pull into a chat is handled by Anthropic, OpenAI, or Google under your agreement with them, as Mav's MCP page puts it. Check which plan your team signs in with, and how that plan stores chats, before anyone connects customer data.

The NAIC Bulletin Is Written for Insurers

In December 2023, the NAIC adopted a Model Bulletin on the Use of Artificial Intelligence Systems by Insurers. It sets governance and risk-management expectations for insurers using AI. According to NAIC testimony from December 2025, more than half of states have adopted the bulletin or similar guidance.

The bulletin is addressed to carriers. Your carriers can still ask appointed agencies how they govern AI tools and oversee vendors. Applicability varies by state and by whether you're a carrier or an agency, so confirm with your carrier, your state DOI, or counsel.

The same goes for privacy and data-security obligations on customer data. Ask your compliance team how they apply before you connect customer records to any AI tool. This article isn't legal advice.

Mav is SOC 2 Type 1 certified with Type 2 pending, and its security practices are documented. For consent-based outreach under the TCPA, see Mav's compliance center.

How Do You Connect an MCP Server to Claude, ChatGPT, or Gemini?

You add the vendor's server URL as a custom connector in your assistant, sign in with your vendor login, and start asking. Setup screens differ by assistant and change often, so follow your vendor's setup guide for the exact steps. The general sequence looks like this:

  1. Confirm your vendor offers an MCP server and which assistants it supports.

  2. Add the vendor's server URL as a custom connector in your AI assistant.

  3. Sign in with your vendor login when prompted.

  4. Confirm which account the connection reaches.

  5. Start with a simple question you already know the answer to.

  6. Check the answer against your dashboard before trusting harder questions.

If the assistant gets your transfer count right on a week you already reported, you know the connection is reading the right account.

Mav's version is short. You paste one server URL into Claude, ChatGPT, or Gemini, sign in with your Mav login and 2FA, and ask in plain English. On Claude Team and Enterprise plans, an owner adds the connector once for the whole organization. ChatGPT connects through developer mode on paid plans that include it, and on Business and Enterprise a workspace admin may need to allow it first. Claude Code takes one terminal command. The Mav MCP server page has the steps for each.

How Do You Pick an MCP Server for Your Agency?

Start with these questions for the vendor. They're practical design checks drawn from the MCP spec and OWASP, and your regulatory obligations still come from your carrier, your state, and counsel.

  • Write access: Is the server read-only by default, and what confirms any action that changes data?

  • Sign-in: Does sign-in use OAuth with least-privilege scopes?

  • Account scoping: Does access match each user's existing permissions and account?

  • Data exposed: Which records reach the assistant, and how is PII handled?

  • Audit logs: Can you see which questions were asked and which data was read?

  • Assistants supported: Which assistants does it work with today?

  • Maintenance: Does the vendor publish a changelog and update the server as the spec changes?

A vendor that answers all seven clearly is one you can take to your carrier or E&O review.

Then match the server to what you're trying to find out:

  • If you're asking about leads Mav worked, like transfers, calls, transcripts, or quotes, connect Mav's MCP server.

  • If the answer lives in your AMS, like policies, renewals, or commissions, ask your AMS vendor whether it offers an MCP server, and keep using its reports until it does.

  • If you want the AI to send messages or update records, look for a server that asks you to confirm each change, and run it past compliance first.

  • If your team uses personal AI accounts, settle which plan they'll use before anyone connects customer data.

  • If nobody will check answers against the source, hold off on questions that drive spend or staffing until someone can.

FAQs

Can I Ask ChatGPT Questions About My Insurance Agency's Lead Data?

Yes, if the system that holds your lead data offers an MCP server that works with ChatGPT. Mav's MCP server connects through ChatGPT's developer mode on paid plans that include it, and it also works with Claude and Gemini.

Is There a Difference Between MCP and an MCP Server?

Yes. MCP is the open standard, the shared rules for how AI assistants connect to outside systems. An MCP server is one vendor's implementation of those rules for its own system, like the server Mav runs for your Mav account.

Is an MCP Server the Same as an API?

They're related. An API gives a developer endpoints to write code against, and an MCP server wraps a system's access in one standard format that any MCP-compatible assistant can use directly.

Is It Safe to Connect Insurance Customer Data to ChatGPT or Claude?

It depends on the server and on your AI plan. Look for sign-in with your own login, least-privilege access, account scoping, read-only defaults, and redaction of sensitive numbers, and check how your plan stores chats. Run the decision past your compliance team before connecting customer data.

Can an AI Connector Change Data in My Insurance Agency Software?

Only if the vendor built actions that change data into its server. Mav's MCP server is read-only, so it can look things up but can't text, call, or change anything in Mav.

Do I Need a Developer to Set Up an MCP Server?

Not for a vendor-hosted server like Mav's: you paste one URL into your assistant and sign in with your Mav login. Building your own server from scratch takes a developer.

Is There an MCP Server for Insurance Agency Lead and Sales Data?

Yes. Mav's MCP server answers questions about the leads, calls, transfers, clicks, and conversions in your Mav account. For AMS data, ask your vendor directly, since offerings are changing quickly. Socotra and Fenris already offer MCP servers on the carrier and data side.

Mav Team

Ready to grow your book?

Mav will make your phone ring.