Connect Your CRM to AI: How Salescadia's MCP Server Works
Connect your CRM to AI assistants like Claude and Cursor with MCP. Ask about your own deals, contacts, and call transcripts — setup takes two minutes.
You can now connect your CRM to AI assistants directly. Salescadia ships a hosted MCP server — an open-standard endpoint that lets Claude, Cursor, or any MCP-capable assistant read your contacts, deals, revenue, meeting summaries, and call transcripts, and answer questions about them in plain language. Setup is an API key and one config line.
This post covers what that unlocks, how the security model works, and the exact setup for the most common clients.
Why Connect Your CRM to AI at All?
Every sales team we talk to has the same pattern: the data is in the CRM, but the questions happen everywhere else. A founder writing an investor update wants this quarter's won revenue. A rep prepping for a renewal call wants a recap of every prior conversation. A manager wants to know which open deals have gone quiet.
Each of those is a five-minute detour today: open the CRM, find the right tab, filter, read, summarize by hand. With an MCP connection, they're one sentence to the assistant you already have open:
- "Which open deals haven't had a touchpoint in two weeks?"
- "Summarize my last three meetings with Acme and draft next steps."
- "What objections did prospects raise about pricing this month?"
- "What's our active MRR right now, and how does the deal funnel look?"
The last two are the interesting ones. Because Salescadia records and transcribes sales calls, the MCP server exposes semantic search across every call transcript — the AI can find the moment a prospect pushed back on pricing even if nobody tagged it, quote it with the timestamp, and pull the full AI meeting summary for context. That's not "CRM fields in a chat window"; that's your team's actual conversations becoming queryable.
How the transcript tools work: they return raw excerpts and let your assistant do the reasoning. Nothing is summarized on our side first, so you get the same quality of analysis your AI produces everywhere else — grounded in what was actually said on your calls.
What the MCP Server Exposes
Eight read-only tools, all scoped to your workspace:
| Tool | What it answers |
|---|---|
search_contacts / get_contact | Who is this person? Full timeline: deals, activities, emails, meetings |
list_deals | Pipeline by status — open, won, lost |
list_tasks | What follow-ups are due, and whose |
get_revenue_summary | Won revenue, active MRR, deal funnel counts |
list_meetings / get_meeting_summary | Meeting history and the AI summary of any recorded call |
search_call_transcripts | Semantic search across every recorded conversation |
Deal statuses come from the same source of truth the Salescadia revenue tab uses — deals are auto-marked won or lost from your connected Stripe, so "how much have we actually collected?" gets a real answer, not a guess. If you want the background on why routing and call data live in one place, the MedLeague case study shows what the underlying dataset looks like in practice.
Setup: Two Minutes, Any Client
- In Salescadia, open Settings → Integrations and create an API key in the Zapier & API Keys card. The read scope is all the MCP server needs.
- Add the server to your assistant with the key in the Authorization header.
For Claude Code, it's one command:
claude mcp add --transport http salescadia https://www.salescadia.com/api/mcp \
--header "Authorization: Bearer mm_live_YOUR_KEY"
For Cursor, Windsurf, or VS Code, it's one entry in mcp.json:
{
"mcpServers": {
"salescadia": {
"url": "https://www.salescadia.com/api/mcp",
"headers": { "Authorization": "Bearer mm_live_YOUR_KEY" }
}
}
}
Custom agents built on any MCP SDK connect the same way — the server speaks standard Streamable HTTP with no session ceremony. The full setup guide has a curl smoke test and troubleshooting.
The Security Model
Bringing an AI to CRM data raises fair questions, so here's exactly how access works:
- Read-only by design. Every tool is a read. Nothing an assistant does through this connection can modify or delete CRM data.
- Workspace-scoped keys. A key sees its own workspace and nothing else — the same isolation our Zapier and Clay integrations use.
- Header-only auth. The endpoint refuses API keys passed in URLs, so credentials never land in request logs.
- Instantly revocable. Deactivate a key in Settings and every AI that used it loses access on the next request.
- Scope enforcement. A write-only key issued to a narrow automation cannot read your CRM through this endpoint.
The one thing to internalize: anything the AI reads becomes part of that AI conversation. Connect assistants your team already trusts with customer data, under whatever data-processing terms you hold with that AI vendor.
Where This Goes Next
Read tools are the foundation. The obvious next layer — which we're building toward — is action with approval: the AI drafts the follow-up email from the call summary, and the rep approves before anything sends. If you want a CRM built for that future instead of bolted onto it, start a trial or see the pricing — the MCP server is included on every plan, with transcript search on Pro and up.
Connect your CRM to your AI
Create a workspace, add an API key, and ask your assistant your first pipeline question — the whole loop takes under ten minutes. The MCP server is included on every plan.
Start Free