Sellers can use sales data inside Claude or ChatGPT by connecting a data source to their AI assistant through MCP (Model Context Protocol). The assistant then pulls what the seller needs on request, such as which companies to work this week and why, without the seller opening another tab. The data tool's own interface becomes optional.
This matters because the hardest part of sales tools is rarely the data. It is getting sellers to use it every day. This post explains the adoption problem, how bringing data to the assistant solves it, and what a seller's day looks like when it works. For the wider case for working without a tool interface, see do you still need a sales tool UI.
Most sales teams have seen this pattern. A rep watches a demo of a new tool and thinks it is great. Then the week fills up with calls, follow-ups, and deals, and the tool goes unopened.
Reps are busy, not careless. A tool that needs its own daily visit competes with everything else on the calendar. Another login, another screen, and another set of menus to learn is easy to postpone. The training that comes with a new tool is also rarely completed, because live work always comes first.
The result is a tool that works well for the few people who use it and does little for the rest of the team. Leaders pay for data that sellers never see.
Many sellers already spend part of their day in an AI assistant (like Claude or ChatGPT). They use it to draft emails, prepare for calls, and summarize notes. If the sales data shows up there, nothing new has to be opened.
With MCP, a data source connects to the assistant once. The seller asks a question in plain language, the assistant calls the data source, and the answer comes back in the same conversation. Syft works this way. A seller can ask which companies are working on the problem they sell against right now and why, and the assistant returns those accounts with dated evidence and a source link for each reason, plus who to contact. The Syft interface stays available for anyone who prefers it, and it is optional. For an overview of what exists today, see the best MCP servers for go-to-market teams.
A common objection is that a new tool means a new training program. With an MCP connection, much of that goes away because the seller can ask the assistant how to use it.
A seller can type, "How do I use Syft to find accounts for this week?" and get a clear answer. The plugins and skills that come with the connection also suggest the next best step, such as turning a list of accounts into a short brief or preparing a first message for review. The seller learns by using it on a real task, in the tool they already know.
Outside context is more useful when it sits next to what your team already knows. The assistant can read your CRM records, call notes, and past deals, then combine them with what Syft returns:
The assistant does this in one place. Nobody has to build a custom integration between each pair of tools. The assistant connects each source once and reads from all of them when it answers. For how this works at the architecture level, see MCP for sales: how to give GTM AI agents real account context.
Many companies have already approved an AI assistant and set rules around it: who can use it, what data it can read, and what it can do. A data source connected through MCP runs inside that setup. Access follows the controls your company already chose for the assistant, and the connection can be reviewed like any other connector. That removes a common blocker: sellers do not need a separate security review for a new standalone app.
Here is a short, illustrative walk-through of one seller's day. It is an example, not a customer story.
Monday, 8:30 a.m. The seller opens Claude and asks, "Who should I work this week and why?" The assistant calls Syft and returns a short list of companies, each with the reason to act now, the date of the evidence, and a link to the source.
Monday, 8:45 a.m. The seller opens two of the sources, confirms the reasons hold up, and asks, "What do we already know about the top account?" The assistant reads the CRM and call notes and replies that a colleague spoke with a different contact last quarter, and summarizes what was said.
Monday, 9:00 a.m. The seller asks for a short first message for the top account, based on the evidence. The assistant drafts it. The seller edits it and sends it.
Wednesday. Before a discovery call, the seller asks, "What is this company working on that relates to our product?" The assistant answers from the dated evidence, so the first questions on the call start from the buyer's actual situation.
Friday. The seller asks, "Which of the accounts I contacted this week replied, and what should I do next?" The assistant combines the CRM updates with the evidence and suggests next steps. The seller decides.
At no point did the seller open another application. The data came to them, and the seller's time went to reading sources, thinking, and talking to buyers.
Questions that work well on day one:
For more on giving an assistant this context, see how do I feed Claude who and why now for outbound.
Syft is the source of outside context in this setup. It follows what companies say and do, and returns the ones working on the problem you solve right now, with a date and a source for each reason. Sellers reach it through MCP in the AI assistant they already use (like Claude or ChatGPT). Syft works in Claude, ChatGPT, and Cursor through MCP.
Better context in, better actions out.
In a custom demo, Syft runs on the problem you solve and shows companies working on it now, with the dated evidence for each. You keep the results whether or not you buy. Book a custom demo.
Yes. With MCP, a data source connects to the AI assistant, and the seller asks questions in plain language. The assistant pulls the answer from the data source in the same conversation.
No. The interface is optional. Sellers can get accounts, evidence, and contacts from the assistant they already use.
Very little. Reps can ask the assistant how to use the connection, and the plugins and skills that come with it suggest the next step.
Yes. The assistant combines what the data source returns with your CRM, call notes, and past deals, without a custom integration between each pair of tools.
The connection runs behind the controls your company already set for its AI assistant. Review it like any other connector.
"Who should I work this week and why?" The assistant returns accounts with dated evidence and a source for each reason.
