Written by sales people who ship, not by Anthropic
Two disclosures up front, because you should know both before you take advice from a website.
Not affiliated with Anthropic. This site is independent. It is not authored, sponsored, reviewed, or endorsed by Anthropic. Claude and Model Context Protocol are theirs. We document how the protocol behaves in sales work, and when we get something wrong, that is on us and not on them.
We are not neutral about one thing. This site is maintained by the team behind Salesgear, which sells sales software and operates an MCP server of its own. That is a commercial interest and you should weigh it. What we can offer in return is that the config, the failure modes, and the workflows here are the ones we actually run. Judge the specifics: the error table on the setup page and the debugging notes are not the kind of thing a marketing page bothers to include.
Who is behind it
Sales practitioners with decades between us, not researchers writing about a category from the outside. We have carried a number, built the systems that missed it, and rebuilt them. Between us that covers the dialler era, the marketing automation era, the sequencer era, and now this one.
That history is the reason this site is structured the way it is. It is not organised around what the technology can do. It is organised around where a quarter actually gets won or lost, because that is the only structure that survives contact with a real pipeline.
The argument
Every meaningful edge in selling for the last forty years came from adopting a tool a year or two before it was obvious. None of them changed what selling is.
| Era | Where the edge was | What did not change |
|---|---|---|
| Print and direct mail | A mailing list nobody else had | Knowing exactly who you are writing to |
| The fax blast | Reaching a desk before the post did | Speed of first contact still wins deals |
| Cold calling floors | Volume plus a script that survived contact | Handling an objection live is still the hardest skill in sales |
| Email and marketing automation | Sequencing and measurement | Follow-up beats brilliance, and always has |
| LinkedIn and social selling | Reaching a buyer who ignored the switchboard | Relevance beats reach |
| Workflow automation, n8n and the rest | Removing the human from the parts that never needed one | A system you can repeat beats a hero quarter |
| Claude, ChatGPT, MCP | Reading and reasoning over context nobody had time to read | Being unknown, unproven, and low on someone else's priority list |
Read the right-hand column again. Forty years of tooling and the hard part has not moved. A stranger has a problem, limited budget, three competing priorities, and no reason to trust you. No model solves that. It just gets you to the conversation faster and better prepared than the rep who did not bother.
What we think this era actually changes
The honest version is narrower than the marketing version. What is genuinely new is that a model can read context no human had time to read: six months of tickets, forty closed-lost notes, a 90 minute transcript, the thread nobody reopened before the call. That is a real edge and it is available today to anyone willing to connect their data.
What has not changed is that this only matters if someone does something useful with the output. The most underrated tool in a room full of AI tooling is a rep who can ask a hard question and sit through the silence afterwards.
The old outbound playbook, the predictable pipeline model that ran B2B software for a decade, does not work the way it did. Volume stopped being a moat the moment everyone got the same tools. What survived from it is the unglamorous part: define who you sell to, write it down, measure the funnel by stage, and follow up more than feels comfortable.
An awkward observation
A large amount of money in this category is spent by people who are measured on having bought the tool rather than on what it produced. Procurement closes, the seat count is reported, and nobody checks in month four whether a single rep changed how they work.
The counter is boring and it is the same one that worked in every previous era. One workflow, one team, one month, with a number attached. If it does not move, stop. That standard is why the pages here list what each workflow needs and where it fails, rather than promising a percentage.
Two kinds of people wrote this
Veterans who have sold through several supposed revolutions and are sceptical of all of them, and operators who adopt anything new the week it ships. Both are half right, which is the useful part.
The veteran is right that the fundamentals do not move. The operator is right that the edge goes to whoever adopts early, and that the window closes fast once everyone has the same tools. Holding both at once means you cannot settle the question by having an opinion. You have to keep testing, and write down what happened.
That is what the log is: a dated record, updated weekly, of what we tested, what changed underneath us, and what we stopped recommending. Every entry names the test or the source behind it, including the ones that make our own earlier advice look wrong.
How to use this site
If you are connecting Claude for the first time, start with the Claude app walkthrough, which needs no terminal. If you want to see what the setup pays for before doing it, the workflows by sales stage is the honest preview. Everything here is free to copy, including into a competitor's enablement deck. The ideas were never the moat.