An MCP CRM is a CRM that AI assistants like Claude and ChatGPT can connect to through the Model Context Protocol (MCP), so the assistant can search, create and update records for you. Some CRMs only let the assistant read; others let it write, with or without your approval for each change.
An MCP CRM is one kind of AI CRM. What an AI CRM is, and the three kinds →
The CRM runs an MCP server that lists the actions an assistant may take, such as “find company”, “update deal stage” or “log a call”. You add that server as a connector in Claude or ChatGPT and sign in.
The assistant then picks the right action for what you ask, and the CRM checks your permissions before it does anything. You don’t install software or write code.
Prep me for my call with Northwind at 2pm.
Log today’s call with Northwind and update the deal.
Draft the follow-up and remind me on Friday if they haven’t replied.
Which deals haven’t moved in 30 days?
What an assistant can actually do depends on the CRM: some connectors only read, and some save changes without asking. The table below shows which is which.
| CRM | Works from Claude | Works from ChatGPT | Assistant can write, not just read | Approval step before changes are saved | Pricing model |
|---|---|---|---|---|---|
| Capable | Yes | Yes | Yes | YesField by field | $29 a month per workspace plus $0.49 per active account. Unlimited seats. Prospects are free. |
| HubSpot | Yes | Yes | Yes | Not stated | Per seat, per month |
| Attio | Yes | Yes | Yes | YesPer action, in the assistant | Per user, per month |
| Pipedrive | YesIn beta | YesIn beta | Yes | Not stated | Per seat, per month |
| Salesforce | YesOn Enterprise Edition and above | YesOn Enterprise Edition and above | Yes | Not stated | Per user, per month |
| Close | Yes | Yes | Yes | ||
| Clarify | Yes | Yes | Yes | Unlimited users; AI features use credits | |
| Breakcold | Yes | Yes | |||
| folk | Yes | YesIn developer mode | Yes | Per user, per month | |
| Day AI | Yes | Yes |
The same rows as our full comparison, checked October 2026 against each company’s own product, help and pricing pages. “Not stated” means we found no public statement either way; a blank cell means we haven’t verified it on the vendor’s own site yet. See the full comparison of CRMs for Claude and ChatGPT →
A CRM with an MCP connector adds assistant access to a product built for typing into forms. An AI-native CRM is built for the assistant to do the admin: data comes in from email, calendar and calls, the assistant proposes updates, and people review them.
Capable is built that way. One MCP connection with 66 tools, each built around a whole job, is the main way to use it; the web app is there for reading and quick fixes.
Both are kinds of AI CRM. What an AI CRM is, and how to choose one →
Model Context Protocol, an open standard Anthropic introduced in 2024 for connecting AI assistants to tools and data. Claude and ChatGPT both support it.
No. You add the connector in Claude or ChatGPT settings and sign in.
It depends on the controls: look for per-change approval, role permissions and an audit log. In Capable, nothing your assistant infers is saved until you approve it field by field, and every change is kept in the audit history.
It depends on what you want the assistant to do. If you want it to do your CRM updates, with your approval and without per-seat pricing, Capable is built for that. If you also need marketing tools, look at HubSpot; if your team mostly works in the CRM’s own app, look at Attio. Our full comparison of CRMs for Claude and ChatGPT covers ten side by side.
Capable keeps the record. You approve what changes.
No card needed. Or talk to us first.