Claude Code is Anthropic's agent that reads files, runs tools and follows written instructions across many steps. For a salesperson, that means account research, prospect lists and first drafts prepared from your own notes and connected systems. This guide covers setup, skills, MCP connections and the approvals to keep.
Claude Code was built for software work, but the same abilities suit repeatable sales research: it reads your notes, calls the tools you connect and follows instructions you write once. You do not need to write code to use it.
We make Buena, which connects to Claude Code, so Buena is the worked example here. Claude Code facts come from Anthropic's documentation, checked on September 28, 2026, and the sources section links to each page.
What Claude Code is, in sales terms
Anthropic describes Claude Code as an agentic coding tool that reads your codebase, edits files, runs commands and integrates with your development tools. Replace the codebase with a folder of account notes and prospect lists, and you have the sales use: Claude works through several steps on its own, using the files and tools you give it.
It runs in the terminal, in IDE extensions, in a desktop app and on the web. The desktop app includes Claude Code, so you do not need to install anything from a terminal. It needs a paid plan: Anthropic lists Pro, Max, Team, Enterprise or a Console account, and the free Claude plan does not include Claude Code.
Sales work it handles well, and what to keep for yourself
| Task | What to hand Claude Code | What you still check |
|---|---|---|
| Account research | A brief on one company: what changed recently and who is likely involved, with a source for every claim | That each source says what the brief claims, and that it is recent |
| List building | A prospect list checked against your written customer profile, with a reason for each row | Rows it marked uncertain, and anyone who should not be contacted |
| First drafts | A short message built from the brief and your notes | Every personal line, the ask and the tone |
| Call preparation | A one-page brief from your notes and the account history | What you know that isn't written down |
| CRM clean-up | Duplicates, missing fields and stale owners in an exported file | Every change before it goes back into the CRM |
Keep three things for yourself: deciding who is worth contacting, anything that sends, and anything that spends money. Claude can reach a wrong conclusion from a thin source, so a brief without links to its sources is a draft, not research.
Set it up in four steps
- Open Claude Code. The desktop app is the easiest start; the terminal and IDE versions work the same way once installed.
- Make a folder for your sales work and add a CLAUDE.md file to it. Claude reads this file for standing instructions: what you sell, who you sell to, and how sources must be cited.
- Connect the systems Claude should read from or act in, using MCP. Start with one.
- Turn work you repeat into a skill, so the same brief comes out the same way every time.
A starting CLAUDE.md
# Sales research instructions
We sell customer-event coordination software to B2B teams.
Good fit: 200 to 2,000 employees, runs recurring customer events.
- Cite a source link for every claim about a company or person.
- Label anything you infer as an inference.
- Never send, publish or buy anything. Prepare drafts for review.
- Keep first messages under 90 words, with one clear question.The company and profile above are a fictional example; replace them with your own.
Turn repeat work into a skill
A skill is a folder with a SKILL.md file that holds the instructions for one kind of task. Claude loads a skill when its description matches what you asked, or you can run it directly by typing a slash and its name. Personal skills live in ~/.claude/skills/ and work in every project; project skills live in .claude/skills/ inside the folder.
---
name: account-brief
description: Research one company before outreach. Use when asked for an account brief or prospect research.
---
1. Find what changed at the company in the last 90 days: funding, hiring, launches, leadership.
2. Name the two or three roles most likely to care, and why.
3. Give one reason to reach out now, tied to a linked source.
4. List the questions you could not answer from sources.
5. End with a table: claim, source link, date.Then ask for “/account-brief Brightwater Freight”. Keep each skill to one job; a skill that tries to do everything is hard to check.
Plugins bundle skills, agents, hooks and MCP servers into one install, through the /plugin command. Anthropic warns that a plugin you install can run code with your user privileges, so install only from sources you trust.
Connect your systems with MCP
MCP, the Model Context Protocol, is how Claude Code reaches other software: a CRM, a data provider, your inbox. Each system publishes an MCP server. You add a hosted server once with its web address, then type /mcp inside Claude Code, pick the server and sign in in your browser.
claude mcp add --transport http <name> <server-url>Adding --scope user makes a server available in all your projects; without it, the server works only in the current folder. Connectors you already added in the Claude apps are available in Claude Code automatically when you sign in with your Claude account. On Team and Enterprise plans, an Owner adds those connectors.
Which servers to add first depends on where your customer data lives. Our guide to MCP servers for sales compares the official Salesforce, HubSpot, Apollo, Clay and Gmail servers. Anthropic's advice applies to all of them: verify you trust a server before you connect it.
Example: from one account to an approved draft with Buena
Buena researches prospects, reveals verified work emails and drafts personal messages, and its MCP server gives Claude Code those abilities. The command is the one shown in your Buena workspace under Settings, then Integrations, then AI Tools:
claude mcp add --transport http --scope user buena https://mcp.buena.ai/mcpStart Claude Code, type /mcp, select Buena and sign in. Then ask for work the way you would ask a colleague:
“Find VP Sales leaders at logistics companies in Germany who recently started a new role. Research the three best fits, with sources.”
“For Maya at Brightwater Freight, reveal her verified work email and draft a first message from the brief. Keep it under 80 words.”
“Put the approved prospects into a draft campaign. Don't activate it.”
The people and companies above are fictional. The connection works under your Buena permissions, and neither Buena connection activates a campaign or sends a message on its own: sending needs an explicit approval, which you give in Buena or by confirming Buena's approve step in Claude Code.
Decide how much Claude Code may do without asking
Claude Code has permission modes. In Manual mode it stops and asks before most actions that edit files, run commands or reach the network. From version 2.1.283, terminal and VS Code sessions start in auto mode, where a separate model reviews each action instead of asking you. Switch modes with Shift+Tab in the terminal, or with the mode selector next to Send in the desktop app.
For sales work, keep the rule simple: let Claude research and draft freely, and make sure you are the one who approves anything that sends, publishes or spends. Buena's send step needs an explicit approval, but your mode decides whether Claude Code asks you or approves it itself. Use Manual mode when you approve campaigns from Claude Code, or approve them in Buena instead.
Claude Code or the Claude app?
Buena also works in the Claude apps on the web and desktop, added as a custom connector. On Free, Pro and Max plans, open Customize, then Connectors, select the plus button, choose Add custom connector and enter https://mcp.buena.ai/mcp. On Team and Enterprise plans, an Owner adds it in Organization settings and members then connect it. Free plans are limited to one custom connector.
Claude can ask for your approval before it uses a connector's tool, and you can set each tool to always allow, need approval or be blocked. Set Buena's approve-and-activate tool to Needs approval, so every send comes back to you. Connectors you add this way also appear in Claude Code when you sign in with your Claude subscription.
Use the app for one-off questions and quick research. Use Claude Code when the work repeats, spans files, or needs standing instructions and skills.
Sources
- Anthropic: Claude Code overview
- Anthropic: Claude Code setup and plans
- Anthropic: Connect Claude Code to tools via MCP
- Anthropic: Skills and plugins
- Anthropic: Permission modes and CLAUDE.md memory
- Anthropic: Custom connectors using remote MCP and using connectors
- Anthropic: How an Anthropic field marketer uses Claude Code to send weekly personalized updates to every sales rep
Claude Code details were checked on September 28, 2026. Claude Code changes often, so confirm commands and plans in Anthropic's documentation. Buena's setup for each AI client is on our MCP page.