Wiring Claude Desktop to your CRM via MCP

What you'll build

By the end of this post, your Claude Desktop will:

1. Accept natural-language prompts like "Find 50 CTOs at Series A fintechs in California, score them on hiring velocity, push the top 10 to my HubSpot."
2. Call TexAu's enrichment, scoring, and CRM-write actions as MCP tools.
3. Return the result with audit trail — what was searched, what got enriched, what was pushed.
4. Run end-to-end without you switching tabs.

This isn't a thought experiment. The MCP server is live at mcp.texau.com/mcp. The Claude Desktop MCP support shipped late 2024 and is generally available. Everything below works today.

Prerequisites

• Claude Desktop, latest version. macOS, Windows, or Linux. (Free Claude.ai web doesn't ship MCP yet — Desktop is required.)
• A TexAu account with an API key. Free tier is fine; MCP works on every plan including free.
• A HubSpot or Salesforce account with admin access to install the connector. (If you don't have one, you can skip the CRM step and stop after enrichment + scoring.)

Total cost to follow this tutorial: $0. Time: 10 minutes.

Step 1 — Get your TexAu API key (1 minute)

Sign in to TexAu. Click your profile → API keys → Create new key.

Name it something memorable like "Claude Desktop." Copy the key — it starts with tex_live_. You'll paste it into Claude Desktop's MCP config in a moment. The key is shown only once; if you lose it, generate a new one.

This key authenticates both the REST API and the MCP server. One key, both rails.

Step 2 — Connect TexAu MCP to Claude Desktop (2 minutes)

Open Claude Desktop. Click the gear icon → Settings → Developer → Edit Config.

A claude_desktop_config.json file opens. If you've never used MCP before, the mcpServers section may be empty. Add the TexAu entry:

{
  "mcpServers": {
    "texau": {
      "url": "https://mcp.texau.com/mcp",
      "headers": {
        "Authorization": "Bearer tex_live_YOUR_KEY_HERE"
      }
    }
  }
}

Save the file. Restart Claude Desktop.

When it reopens, you'll see a small MCP indicator at the bottom of the chat input — confirming the TexAu server is connected. Click it to see the tool catalog Claude can now call.

The tool catalog includes all 249 TexAu actions: texau.enrich.email, texau.enrich.company, texau.score.lead, texau.crm.push, and many more. They're discoverable; Claude reads the catalog and knows what's available.

Step 3 — Run your first agent prompt (90 seconds)

In Claude Desktop, type:

Find 5 CTOs at SaaS companies with 50–200 employees who raised Series A in the last 12 months. Enrich their LinkedIn URLs and email addresses, score them on hiring velocity, and show me the results.

Claude does three things:

1. Plans the call sequence. It introspects the TexAu tool catalog, picks the right actions (texau.search.companies, texau.search.executives, texau.enrich.profile, texau.enrich.email, texau.score.lead), and tells you what it's about to do.
2. Asks for approval. Claude Desktop shows you the plan and waits for you to confirm. This is the "agent acts on my behalf only when I approve" pattern — important because the actions cost credits.
3. Runs the chain. On approval, Claude calls each tool in sequence, accumulating the result.

About a minute later, Claude returns a list of 5 CTOs with names, companies, LinkedIn URLs, verified emails, hiring-velocity scores, and the top three reasons each scored that way.

You spent zero seconds on a Clay canvas. You wrote zero lines of glue code. You typed one sentence.

Step 4 — Add the CRM-write step (3 minutes)

Now let's wire HubSpot or Salesforce as the destination.

In TexAu, go to Integrations → Connect HubSpot (or Salesforce). OAuth flow: two clicks. Field mapping: TexAu auto-detects HubSpot's standard contact properties, so unless you have custom fields, you can accept the defaults.

Back in Claude Desktop, retry the prompt with the CRM-write step:

Find 5 CTOs at SaaS companies with 50–200 employees who raised Series A in the last 12 months. Enrich their LinkedIn and emails, score on hiring velocity, and push the top 3 to my HubSpot as new contacts in the "AI-sourced leads" list.

Claude plans:

• Same first 5 actions as before.
• Then texau.crm.push with crm: hubspot, list: AI-sourced leads, records: top_3_by_score.

It asks for approval. You approve. About 90 seconds later, Claude reports: 3 contacts pushed to HubSpot, with the property values populated and a TexAu-side activity log entry.

Open HubSpot in another tab. The 3 contacts are there, in the right list, with the score visible as a custom field, and a timeline entry that says the contacts came from TexAu via Claude Desktop on today's date.

Step 5 — Save the prompt as a daily routine (90 seconds)

The pattern above is a one-shot. The interesting part is making it a routine.

In Claude Desktop, you can save the prompt as a project. Tag it daily-prospect-flow. Add a system prompt that contextualizes it for your team's ICP and CRM conventions.

Now every morning you can open Claude Desktop, click the project, type "run today's flow," and the agent re-runs the chain with the current data. Different funding announcements, different new hires — the agent picks up whatever's fresh.

If you want to remove yourself from the loop entirely, set up a TexAu scheduler (covered in the waterfall tutorial) that runs the same chain via the API at 4 AM. The agent + the scheduler are two ways to express the same pipeline. Pick what fits your workflow.

Why this works (and why it didn't a year ago)

A year ago, this tutorial wouldn't have worked. You'd have needed:

• A Custom GPT in OpenAI with custom Actions wired to the TexAu API.
• A separate Cursor configuration for the developer half of your team.
• A hand-rolled webhook bridge between your AI tool and your CRM.
• A maintenance burden every time TexAu, HubSpot, or Anthropic changed an interface.

MCP collapses this to: one config file, one tool catalog, agnostic to which agent runtime you use. Claude Desktop today, Cursor tomorrow, your own LangGraph agent next month — they all read the same MCP catalog, call the same tools, and produce the same results.

This is the architecture argument from Why every GTM team will run an MCP server by 2027, reduced to a 10-minute setup.

What to try next

Once the basic flow works, the interesting expansions:

Add prompt drafting

Add an AI Column inside TexAu that drafts a 50-word personalized opener per contact, referencing their hiring velocity. The agent calls it as part of the chain. Outbound personalization without you writing each one.

Push to outbound, not just CRM

Replace the texau.crm.push step with texau.outbound.push to Smartlead or Instantly. Claude assembles the list, drafts the personalization, and queues the campaign — you review the campaign before sending.

Subscribe to signals

Set up a TexAu webhook for "score crossed 80 + executive hire detected." Wire it to a Slack channel. Now Claude Desktop also gets pinged in real time when any account in your list passes the threshold — not just when you ask.

Limitations to know about

• Claude Desktop's MCP support is generally available but evolving. The exact UI for managing servers and approving plans may shift slightly across versions.
• Approval is per-call by default. For high-volume workflows, configure batched approval in Claude Desktop's settings, or move the flow to a scheduler.
• Rate limits apply. TexAu's standard rate limits apply to MCP calls the same as REST API calls. If you're running thousands of calls per minute, talk to sales about lifted limits.

None of these are dealbreakers. They're sharp edges of a still-young protocol that will smooth over the next year.

TL;DR

• TexAu MCP server: https://mcp.texau.com/mcp
• Auth: Bearer token from your TexAu API keys page
• Works with: Claude Desktop, Cursor, Continue, any MCP-aware client
• Free tier: yes, MCP works on every plan including free
• Use case: end-to-end GTM workflows from a single chat prompt

If you've been waiting for the moment where "AI agent for GTM" stops being a slide deck and starts being something you can use Tuesday afternoon — this is that moment.

Start free. Read the MCP docs. See the full action catalog.