CRM MCP Server: How to Run Your CRM from Claude or ChatGPT
With a CRM MCP server, you can find leads, start outreach and check deals by asking Claude or ChatGPT. Here is how it works and what to check before you connect.
A CRM MCP server is a connector that lets an AI assistant such as Claude or ChatGPT read and update your CRM through the Model Context Protocol (MCP). Once it is connected, you ask in plain language, for example "which deals had no activity this week?", and the assistant calls the CRM's tools to answer or act. Both Claude and ChatGPT can ask for your approval before a tool changes data.
What is MCP, in plain words?
MCP (Model Context Protocol) is an open-source standard for connecting AI applications to outside systems such as databases, business tools and workflows. The official MCP site compares it to a USB-C port for AI apps: one standard way to plug in, instead of a custom integration for every tool. Anthropic open-sourced MCP in November 2024.
According to modelcontextprotocol.io, MCP has three participants:
- Host: the AI app you use, such as Claude or ChatGPT.
- Client: the part inside the host that keeps a connection to one server.
- Server: the program that offers data and actions, such as your CRM's MCP server.
A server can offer three kinds of things: tools (actions the AI can run, such as "create a lead"), resources (data the AI can read) and prompts (reusable templates). For a CRM, tools do most of the work.
What does a CRM MCP server do?
A CRM MCP server turns CRM actions into tools that an AI assistant can discover and call. The assistant asks the server which tools exist, picks the right one for your request, sends the inputs and shows you the result. The server decides what is allowed, so the assistant can only do what those tools permit.
Typical CRM tools include:
- Search contacts, companies, leads and deals
- Create or update a record, such as moving a deal to a new stage
- Add a task or a note
- Find new leads that match a description
- Start an outreach campaign for a list of leads
MCP servers can run on your own computer or remotely. The MCP documentation explains that local servers usually talk to the AI app through standard input/output (stdio), while remote servers use Streamable HTTP and can serve many users. For a cloud CRM, you will normally connect to a remote server.
What can you ask your CRM from Claude or ChatGPT?
You can ask for anything the server's tools support. Most requests fall into a few groups: finding leads, starting outreach, checking your pipeline and preparing for calls. Phrase requests the way you would brief a colleague, and name the CRM when you have several connectors turned on.
| Job | Example request |
|---|---|
| Find leads | "Find logistics companies in Saudi Arabia that run their own warehouses, and add their heads of operations as leads." |
| Start LinkedIn outreach | "Start a LinkedIn campaign for the 25 leads I added today, using our warehouse audit message." |
| Start email outreach | "Write a first email to each lead from yesterday's event and show me the drafts before sending." |
| Check the pipeline | "Which deals in proposal stage have had no activity for 10 days?" |
| Prepare for a call | "Summarize everything we know about this account before my 3 pm call." |
| Clean up data | "List leads that have no company attached." |
The MCP server for Polluxa CRM works this way. From Claude or ChatGPT, you can find leads, start LinkedIn and email outreach, and search deals without opening the CRM. Outreach then runs through the same agents, such as the LinkedIn Agent, and replies land back in the CRM.
How do you connect a CRM MCP server to Claude or ChatGPT?
In general, you add the server as a connector in your AI app, sign in to your CRM account, and turn the connector on in a conversation. Menus change often, so follow the current help pages from Anthropic and OpenAI, plus your CRM vendor's own instructions.
Here is what each AI provider's documentation says today:
- Claude. Anthropic's help center says custom connectors using remote MCP are available on Free and paid Claude plans, with Free users limited to one custom connector. On Team and Enterprise plans, an owner adds the connector for the organization, and each member then connects with their own account. You turn connectors on for each conversation.
- ChatGPT. OpenAI's developer documentation describes adding a custom MCP server on the web with its URL and an authentication method (OAuth or no authentication). It supports servers that use SSE or streaming HTTP, and write actions require confirmation by default.
For Polluxa, connect the Polluxa MCP server from your Polluxa workspace. If you want help with setup, book a demo and ask the team to go through it with you.
What security questions should you check before connecting?
Check who runs the server, how it signs you in, what each tool can change, and whether your AI app asks before it acts. Anthropic and OpenAI both tell users to connect only to MCP servers they trust. OpenAI warns that an untrusted server could take information you share or push the assistant to misuse tools, such as changing or deleting data.
Work through these questions before you connect a CRM:
- Who built and hosts the server? Prefer your CRM vendor's official server over unknown third-party code.
- How do you sign in? The MCP documentation recommends OAuth for getting access tokens. Per-person sign-in beats one shared key: Anthropic notes that when members connect individually, Claude only reaches data each person can access.
- What can each tool do? Ask for the list of tools. Separate read-only tools from tools that create, change, delete or send.
- Does the app ask before it writes? OpenAI says write actions require confirmation by default in ChatGPT. Anthropic advises choosing "Allow always" only for tools you trust to run unsupervised.
- What about prompt injection? Anthropic warns that malicious servers may hide instructions meant to trigger actions you didn't intend. Be extra careful with any request to export data or send messages in bulk.
- Is there a record of what happened? The MCP specification says servers must validate inputs, apply access controls and rate-limit tool calls, and that clients should log tool use for audits. Ask where you can see what the AI did in your CRM.
- Who on your team may connect? Decide which roles can link the CRM to an AI app, and which tools each role gets.
Example: a 3-person founding team in Toronto
Picture a 3-person founding team in Toronto that sells scheduling software to dental clinics. The sales lead starts each day in Claude: "Which deals need a follow-up today?" Claude searches the CRM and lists them. She asks for a short follow-up draft for each, edits two and approves the rest. Then she asks for ten more clinics in Ontario that look like their best customers and adds them to a LinkedIn campaign. She approves each action that changes data instead of choosing "Allow always," and everything is saved in the CRM.
What are the limits of running your CRM from an AI assistant?
An AI assistant is a fast way to ask questions and run routine actions, but it doesn't replace the CRM's own screens. It can misread a vague request, it can only use the tools the server offers, and its answers are only as good as your CRM data. Bulk changes deserve extra care.
- Use the CRM itself for pipeline boards, reports and settings.
- Check the result of any bulk action before you approve it.
- Keep requests specific: name the stage, date range, list or campaign.
- Remember that channel rules still apply. LinkedIn's User Agreement, for example, prohibits unauthorized automated methods for adding contacts or sending messages, so keep outreach personal and modest.
FAQ
No. If your CRM offers a hosted MCP server, you add it as a connector in Claude or ChatGPT and sign in. Building your own MCP server is developer work, but using one is not.
In principle, yes. MCP is an open standard, and both Claude and ChatGPT support remote MCP servers. Check your CRM vendor's instructions for each app, since sign-in and supported features can differ.
It sees what the server's tools return, within the access of the account you sign in with. Per-person sign-in limits the assistant to what that person can already see. Turn off tools you don't need in a conversation.
An API is a set of endpoints that developers call from code. An MCP server describes its tools in a standard format, so an AI app can discover them and decide which one to call. An MCP server is often built on top of a product's existing API.
It can be, if a person approves each campaign and reviews the messages. Keep approval prompts on for any tool that sends messages, and start with small lists.
Want to see it on your own pipeline? Book a demo of Polluxa CRM and its MCP server.
More Blogs
No More Blogs