Every sales prompt worth running needs data the model cannot see: the deal history, the last four emails, what the account actually uses, who the new VP of Engineering is. Model Context Protocol is the wire that hands those over. This page lists the servers worth connecting, the config that connects them, and the workflows that pay for the setup time.
An MCP server exposes a small set of tools to a model. Claude sees the tool names and descriptions, decides which to call, calls them, and reads the result. You are not writing an integration. You are pointing Claude at a server and letting it decide when the tool is relevant.
The practical effect for a rep: the difference between “here is a generic email about our product” and “here is an email that references the two support tickets they filed last month and the integration they asked about on the last call.” Same model. Different access.
If you want the longer strategic write-up rather than the config files, Salesgear's guide to claude mcp for sales covers the org-level version: what to connect first, who owns permissions, and how to keep a rep from handing the model more scope than the role needs.
Ranked by how often the data comes up in a real sales prompt, not by how interesting the server is.
Contact and company search, email and mobile enrichment, and deep research on a named prospect. The research tool returns a structured brief rather than a paragraph, so the model can quote specifics instead of paraphrasing.
Read and write contacts, companies, deals, and engagements. Useful for pulling a deal's full activity history into a call-prep prompt.
SOQL queries, record CRUD, and metadata reads. Give the model a read-only profile first and widen scope once you trust the prompts.
Thread search, message reads, draft creation, and calendar reads. Draft creation matters more than send: a rep reviews before anything leaves.
Channel and thread search, message reads, and posting. Scope it to deal channels rather than the whole workspace.
File search and content reads across Docs, Sheets, and Slides. Point it at a shared drive that holds proposals, security answers, and pricing sheets.
| Task | Without MCP | With MCP |
|---|---|---|
| Call prep | Open CRM, scroll activity, skim two threads, check the notes doc | One prompt, model reads all four sources, returns the brief |
| Account research | Paste a company website into chat and hope | Enrichment tool returns funding, headcount, tech, hiring signals |
| Follow-up draft | Rewrite a template from memory of the call | Model reads the thread, drafts, leaves it in your drafts folder |
| Pipeline review | Export to a sheet, pivot, eyeball | Ask the question, model queries the CRM and answers it |
| CRM hygiene | Nobody does it | Model writes the call summary back to the record after the meeting |
None of this is speculative. Each row maps to a workflow on the workflows page, with the prompt that runs it. The CRM-write rows in particular are covered in more depth in this walkthrough of five sales workflows to automate with Claude and your CRM.
A model with sixteen servers attached picks the wrong tool more often and burns context describing tools it will never call. Start with two. Add a third when a prompt visibly fails for lack of data.
Give read-only scopes first. Move to drafts rather than sends, and to a staging CRM field rather than the record, until you have watched a week of output.
MCP gives access, not judgment. If the qualification criteria live in a teammate's head, the output is generic regardless of how many servers you attach. Write the criteria down somewhere the model can read, then attach the server that reads it.