MCP
Best MCP servers for sales
Search for the best MCP servers for sales and almost everything you find is written for whoever administers the CRM. A RevOps lead wiring an assistant into Salesforce for a team of forty has a real problem, and the lists are a fair answer to it. It is just not the problem most closers have, which is that the numbers deciding their month live in four places and get updated late on a Friday.
This is the same question from a closer's seat. What each server lets an assistant do, four tests worth applying before you connect anything, and the part none of them solve.
Key takeaways
- An MCP server is an access layer over a product you already have. It moves nothing and stores nothing of its own.
- Judge one on four things: read versus write, whose permissions it inherits, what setup costs you, and whether it holds the data you are actually judged on.
- The official CRM servers are genuinely good at pipeline. None of them is built around one person's dials, objections and commission.
- Connecting an assistant does not make you log anything. The habit is still the hard part, and no protocol fixes that.
What an MCP server actually is
The Model Context Protocol is an open standard for connecting an AI assistant to an external system. A vendor publishes a server, that server exposes a named list of tools, and a compatible client such as ChatGPT, Claude or Cursor can call them on your behalf after you sign in.
Two consequences follow, and both matter more than the protocol itself. The first is that the assistant can only do what the vendor decided to expose. If a tool for something does not exist, no phrasing will conjure it. The second is that you are signing in as yourself, so the server inherits your existing permissions rather than granting new ones. That is a security property worth understanding rather than a limitation to work around.
How to judge the best MCP servers for sales
Read, or read and write
A read-only server answers questions. A server with write tools also changes things, which is the difference between an assistant that reports your pipeline and one that maintains it. Both are useful and they are not the same purchase, and the second one deserves more thought about what happens when the assistant misunderstands you.
Whose data, and under whose permissions
Every server in this list authenticates you individually rather than handing over a shared key, and the good ones say so explicitly. Pipedrive documents that an assistant can only work with information your own user may already see and edit. Salesforce documents per-user OAuth with tools that respect field-level security and sharing rules. Read that section before anything else, because it tells you what an assistant would expose if you asked it a careless question in front of somebody.
What setup costs you
Most of these are a pasted address and a sign-in. Salesforce is the outlier in this set: its hosted servers are turned on and configured in the org first, which is the right design for an enterprise and means a rep cannot simply connect one on a Tuesday afternoon.
Whether it holds the thing you are judged on
This is the test that separates a closer from a sales operations lead, and it is the one the listicles skip. CRM schemas are built around accounts, deals and pipeline stages, because that is what a manager forecasts from. A closer is judged on dials made, meetings held, offers presented, objections heard and commission earned, and most CRMs hold that badly. An MCP server over a CRM inherits the CRM's shape. That gap is what MCP sales tracking is meant to fill, and it is a different job from putting an assistant in front of a pipeline.
The sales MCP servers worth knowing
Everything below was checked against the vendor's own documentation while writing this, and nothing is repeated from a third-party roundup. Where a capability is not listed here, it is because that vendor does not document it plainly, not because it does not exist.
| Server | What an assistant can do with it | Who it is really for |
|---|---|---|
| Salesforce Hosted MCP Servers | Read, create, update and delete records with tools that respect field-level security and sharing rules. Apex invocable actions, Flows and named queries can be exposed as tools without writing connector code. Per-user OAuth. | Admins and RevOps inside a Salesforce org. Configured centrally, then connected by users. |
| HubSpot MCP server | Read and write access to CRM data including contacts, deals and engagements. HubSpot also runs a second, local server aimed at its developer platform rather than at sellers. | Teams already on HubSpot, plus developers building on it. |
| Pipedrive MCP | Search deals, contacts, organisations and leads, retrieve pipeline detail, and create or update deals, contacts and activities. OAuth setup, and actions stay inside the permissions your own user already has. | Pipedrive users who want their pipeline reachable from ChatGPT or Claude. |
| Attio MCP | Search, read, create and update records across people, companies, deals, tasks and notes. Attio documents that read operations are auto-approved while write operations ask for confirmation. | Attio workspaces. The clearest published position on read versus write of anything here. |
| Close MCP | Close publishes an official server, described on its own integrations page as a standardised interface letting any compatible AI model or agent reach your Close data. | Close CRM users. |
| CalcuCloser MCP | Read dashboard totals and goal progress, performance for a week, month or quarter against the period before it, objections and win and loss reasons, offers, pay schedules, upcoming paydays, products and recent sales. Log a sale, a day of calls, or a full sales conversation with its outcome. | An individual closer or setter tracking their own activity and commission. |
The pattern is hard to miss once the table is in front of you. Five of the six are servers over a team CRM, and they are aimed at the person who owns that CRM. That is not a criticism — it is who bought the software. It does mean that a closer connecting one of them ends up with an assistant fluent in pipeline stages and silent on the numbers they actually get paid against.
What none of these fix
An assistant with a write tool still only knows what you tell it. Connect every server in the table and the call you did not mention is still missing from every one of them. This is the same failure that ends most spreadsheet habits, described in commission tracking without spreadsheets, and the protocol does nothing about it. What changes is the cost of the habit: a sentence at the end of a call rather than a form at the end of a week. That is a real improvement and it is smaller than the marketing around MCP implies.
Write tools also deserve a moment of caution. An assistant that misunderstands "book that for Thursday" can create a record you then have to find and unpick. Ask it to confirm details before it writes, and prefer servers that make the read and write boundary explicit rather than leaving you to infer it.
And an MCP server does not tell you which numbers matter. It fetches whatever you name. Deciding what to watch is still the harder half of the problem, which is the argument in leading vs lagging sales indicators and the reason how to track sales performance is worth settling before you connect anything at all.
Where CalcuCloser fits
CalcuCloser by Velisi runs an MCP server at calcucloser.velisi.io/mcp, with instructions for ChatGPT, Claude, Claude Code and generic clients on the connect page. It is the one server in the table built around a single seller rather than a team pipeline: an assistant can read your dashboard totals, your performance against the previous period, your objections and win and loss reasons, your offers, pay schedules and upcoming paydays, and it can log a sale, a day of calls or a full conversation with its outcome, creating the deal, the commission and the payment schedule the way the app does.
It runs as you, on your own data, and it cannot see anyone else's numbers, change your billing or delete anything. It also will not make you log anything you would otherwise skip. If the reason your tracking is thin is that you did not want to open a spreadsheet, telling an assistant instead is a genuine improvement. If the reason is that you would rather not look at the month, no protocol is going to help with that. See the tracker.
Frequently asked questions
What is an MCP server in a sales context?
It is a service that exposes a defined set of tools an AI assistant can call on your behalf. In sales that usually means a CRM publishing tools like search records, create a deal or update an activity, so an assistant such as ChatGPT or Claude can reach live data instead of guessing from what you paste into the chat.
Which sales tools have official MCP servers?
As of September 2026, Salesforce, HubSpot, Pipedrive, Attio and Close all publish official MCP servers documented on their own sites, alongside CalcuCloser's server for individual sales tracking. Plenty of unofficial community servers exist too, and those carry a different trust question because nobody at the vendor maintains them.
Can an AI assistant write to my CRM through MCP, or only read?
It depends on the server. Salesforce, HubSpot, Pipedrive, Attio and CalcuCloser all document write tools as well as read tools. Whether a write happens without you seeing it depends on the client and the server: Attio, for example, documents that reads are auto-approved while writes ask for confirmation.
Do I need a developer to set up a sales MCP server?
For the hosted ones, generally not — Pipedrive, Attio, Close and CalcuCloser all connect through an OAuth sign-in and a pasted server address. Salesforce is the exception in this set: its hosted servers are configured in the org by an admin before any rep can connect a client to them.
Does an MCP server replace my CRM?
No. It is an access layer, not a system of record. The data still lives in the underlying product, the permissions are still that product's permissions, and anything the vendor has not exposed as a tool remains invisible to the assistant however you phrase the question.
What should a closer look for that a RevOps lead would not?
Whether the server stores the thing you are actually judged on. Team CRM servers are built around pipeline hygiene and reporting for a manager. A closer usually wants dials, held meetings, objections and commission, and most CRM schemas hold those badly or not at all.