Claude Desktop CRM: Connect Claude to Your CRM in 5 Minutes

What Claude Desktop is and why it matters for CRM

Claude Desktop is Anthropic’s native desktop app for Claude. It runs on macOS and Windows, supports MCP (Model Context Protocol) for external tool integration, and has become a serious productivity environment for technical and non-technical users alike.

For CRM use specifically, Claude Desktop is interesting because it’s where a lot of revenue-team workflows actually live. Reps draft emails in it. Ops leads run analysis in it. Founders use it for everything from outbound personalization to pipeline review. Connecting your CRM to Claude Desktop means those workflows can happen against your real data, not pasted-in snippets.

The technical enabler is MCP. Anthropic introduced it in late 2024 as an open standard for AI-to-tool communication. Claude Desktop was the first major MCP client, and it supports any MCP-compatible server. CRMs that ship MCP servers are accessible through Claude Desktop natively. CRMs that don’t are not.

What you can actually do with Claude Desktop + Conduyt

A few specific workflows that work today:

“Show me my pipeline.” Claude calls conduyt_dashboard_summary and returns deal counts, stage breakdown, total value, and what’s stalled. Replaces the dashboard tab for most quick questions.

“Reassign all deals stuck in proposal for more than 14 days to Maya.” Claude calls conduyt_deals_list with filters, then conduyt_deals_bulk_assign. Three seconds, one sentence, twelve deals reassigned. The audit log records exactly what happened.

“Draft a re-engagement email to leads who haven’t responded in 30 days, then enroll them in the warm-leads drip.” Claude calls conduyt_contacts_list with the date filter, generates the email through conduyt_compose_email, then calls conduyt_email_sequences_enroll. You approve before anything sends.

“Summarize my Q3 pipeline performance and compare to Q2.” Claude pulls the relevant data, runs the analysis in its own context, returns a written summary. The query that used to take an analyst a half day.

“What changed on the Harborview deal this week?” Claude calls conduyt_deals_history and walks you through the audit trail in plain English.

“Help me prep for my 10am with Ridgeline Co.” Claude calls conduyt_contacts_get, conduyt_companies_get, and conduyt_activities_list, then drafts a pre-meeting brief.

These aren’t demos. They’re what Conduyt customers do daily once they connect Claude.

How the integration compares to other CRM + Claude options

A few honest comparisons:

HubSpot + Claude. HubSpot ships a Claude connector that gives Claude read-only access to standard HubSpot objects (contacts, companies, deals, tickets, invoices, products). The integration is genuinely useful for analytics and prep work. The limit: read-only. Claude can’t update records, change deal stages, draft emails into HubSpot, or trigger automations. It can answer questions about your CRM, but it can’t act on your CRM.

Salesforce + Claude (via Agentforce or MCP). Salesforce supports Claude connections through Agentforce or via its emerging MCP support. The capability is real but premium-gated, often requiring higher-tier Salesforce plans plus per-action AI costs. Setup typically requires a Salesforce admin or consultant.

Zoho + Claude. Zoho MCP supports both read and write through OAuth. Works across Zoho CRM, Mail, Calendar, and other apps in the Zoho One suite. Strong option if you’re already a Zoho customer.

Pipedrive, Close, Salesflare, etc. Most don’t have native Claude integrations. Some have community-built MCP servers of varying quality. Setup requires more engineering work.

Open-source + Claude. Relaticle ships a 30-tool MCP server and is designed for exactly this workflow. Self-hosted means you own the data; you also own the maintenance.

Conduyt + Claude. 104 tools across 20 modules. Full read and write. Native first-party support, not a community port. The MCP server is maintained by us, runs on our infrastructure, and ships with the same auth and audit guarantees as the rest of Conduyt. No premium gating. The $299/mo Conduyt plan includes it.

The pattern: most CRMs are either read-only or premium-gated when it comes to Claude. Conduyt is the option built for full Claude-driven workflows at a flat price.

Setup: actually five minutes

The complete setup for Claude Desktop:

Step 1. Install Claude Desktop from claude.ai if you haven’t already.

Step 2. Sign up for a Conduyt trial. Generate an API key from your account settings (cdy_ prefix).

Step 3. Get the conduyt-mcp package. Available on request during your trial; included in your account dashboard for paid accounts.

Step 4. Open your Claude Desktop config file. On Mac, it’s at ~/Library/Application Support/Claude/claude_desktop_config.json. On Windows, at %APPDATA%\Claude\claude_desktop_config.json.

Step 5. Paste this in (replace the API key with yours):

{
  "mcpServers": {
    "conduyt": {
      "command": "npx",
      "args": ["tsx", "src/index.ts"],
      "cwd": "~/path/to/conduyt-mcp",
      "env": {
        "CONDUYT_API_URL": "https://conduyt.app",
        "CONDUYT_API_KEY": "cdy_your_key_here"
      }
    }
  }
}

Step 6. Restart Claude Desktop. Look for the MCP indicator in the bottom-right of the chat input. If you see it, the 104 tools are live.

Step 7. Test it. Ask Claude “how many open deals do we have?” Claude should call conduyt_deals_list and answer.

That’s the entire flow. The hardest part is opening the config file.

For Claude Code (terminal-based) users, the same JSON works in your Claude Code MCP config. Same tools, same auth, same audit log, different interface.

What about Claude Code?

Claude Code is Anthropic’s terminal-based agent for development workflows. It uses the same MCP system as Claude Desktop, which means everything on this page applies to Claude Code too.

For RevOps teams or technical founders who want CRM operations from the command line, Claude Code + Conduyt is a fast workflow. Run claude in your terminal, ask “show me deals stalled in proposal,” get an answer in your shell. The 104 tools are available in both contexts.

The setup is the same JSON config, in a slightly different file location. The developer docs cover the specifics.

Safety and access control

A few specifics for the security-minded:

Scoped API keys. Generate keys with read-only, write, or admin scope. Restrict to specific modules (“contacts and deals only, no users”). Limit by IP if needed. Expiration dates if you want belt-and-suspenders.

Dry-run mode. For any mutation, append ?dry_run=true to see what would happen without committing. Useful when Claude proposes a bulk action and you want to verify before approving.

Confirmation tokens. Destructive actions (bulk delete, mass reassign) require a 5-minute-TTL handshake. Claude proposes, you confirm with a separate call. A single hallucinated DELETE can’t cascade.

Audit log. Every Claude action is logged: which key, which tool, what payload (PII-masked), when, from where. 90-day retention. Exportable to your SIEM.

Rate limits. Default 120 requests/minute per key. Catches runaway loops before they become real problems.

The AI Safety page covers all of this in more depth.