How to connect your AI to your CRM

You connect your AI to your CRM by exposing the CRM over the Model Context Protocol (MCP) and pointing an MCP-compatible client — Claude, Cursor, or similar — at that endpoint with a scoped token. Once connected, the AI reads your live records and calls typed tools to create or update them, instead of sitting in the corner as a read-only chat bubble. The connection takes minutes; the important decisions are what the AI is allowed to read, what it’s allowed to write, and who approves the writes.

This guide explains the two ways “AI + CRM” is usually done, why the MCP approach is the one that does real work, and how to keep it safe.

Why “connect AI to your CRM” usually means a read-only chatbot

Most CRMs that advertise AI ship a chat bubble: a generic model with a system prompt that can answer questions about your data and little else. It is bolted on, tied to the vendor’s model choice, and — because giving a chatbot real write access safely is hard — usually read-only.

That is fine for “summarise this account,” but it is not what people mean when they want their AI working the CRM: creating the follow-up task, logging the deal, drafting the proposal grounded in real pipeline data. For that, the AI needs genuine read and write access, and you need to control both.

The better pattern: expose the CRM over MCP

The Model Context Protocol inverts the bolted-on model. Instead of the CRM embedding an AI, the CRM exposes itself to whatever AI you bring. An MCP-compatible client connects to one endpoint and gets two kinds of capability:

• Tools — actions the AI can call, each with a typed schema, like sstm_crm_add_contact or sstm_crm_add_deal.
• Resources — read surfaces addressed by URI, like sstm://crm/deals (your live pipeline) or sstm://project/{id}/context (everything about one project).

Because MCP is an open protocol, the same client that works with your CRM works with every other MCP server it’s configured for — there’s no bespoke integration to build per tool.

What your AI can do once it’s connected

Here is the shape of a real session against an MCP CRM surface: the AI grounds itself in live data, then proposes an action.

Step	What the AI does	How
Read	Pulls your open deals	reads sstm://crm/deals
Read	Loads a company’s full context	reads sstm://project/{id}/context
Act	Creates a new contact	calls sstm_crm_add_contact
Act	Logs a deal against a company	calls sstm_crm_add_deal

In sSystm the CRM is one domain of a larger MCP surface — 29 tools and 35 resources spanning contacts, deals, projects, calendar, documents and the design system — so the same connected AI can move from “summarise this account” to “create the project and draft the kickoff doc” without leaving your workspace.

Keeping it safe: human-in-the-loop writes

Read access can be immediate — an AI querying your pipeline changes nothing. Writes are where discipline matters. The safe pattern is simple: the AI proposes, a human approves.

In sSystm’s agent mode, every write is staged as pending. A person sees exactly what the AI wants to do, approves or rejects it action by action, and every run — approved or not — is recorded in an Agent Log the whole team can audit. You also choose which tools are exposed in the first place, so the AI can only ever act within a surface you defined. Read more about the mechanics in the security and data model.

How to connect your AI to your CRM

The concrete steps, using sSystm as the example:

1. Generate a token. Open the MCP module in your workspace and create a scoped token for your AI client.
2. Point your client at the endpoint. In Claude Desktop, Claude Code, claude.ai (via a custom connector) or Cursor, add the MCP endpoint with that token.
3. Ground it in real data. Ask something concrete — “summarise my open deals,” “which projects are stalled” — to confirm the read surface works.
4. Let it act, with the gate on. Ask it to create a contact or log a deal; review the pending action; approve. The record lands in your CRM.

Because sSystm is BYOC, every read and write your AI makes lands in the database on your own cloud account — the same ownership guarantee that applies to your team applies to your AI. Explore the full MCP story for the complete tool and resource map, or the module catalogue to see everything a connected AI can reach.