By Mopsify | Marketing operations for modern B2B teams

Most of what a marketing ops analyst does day to day is pattern-matching work: pull data from a system, reshape it, write something (a report, an email, a list of next steps), and push it back. That is exactly what an LLM with direct tool access is good at. Instead of copy-pasting between browser tabs and spreadsheets, Claude can reason about your data and act on your systems through structured calls.

At Mopsify, we help B2B marketing teams get more out of their tech stack, and connecting Claude to HubSpot and Salesforce has become one of the highest-leverage things we set up for clients. Here is what it unlocks and how to do it.

First, a quick glossary

MCP (Model Context Protocol) – is an open standard, originated by Anthropic and now widely adopted, that defines a common way for AI models to discover and call tools and data sources. Think of it as an AI-native API: a standard socket that any AI client (Claude, Cursor, and others) can plug into, and any system (HubSpot, Salesforce, Slack, GitHub) can expose itself through, without custom integration code for every pairing.

Agentic – simply means the AI doesn’t just answer a question. It plans multiple steps, calls tools, checks results, and keeps going until the task is done.

How to connect each platform

HubSpot: an official, no-code connector

HubSpot exposes its CRM and marketing functionality as MCP tools built for an LLM to interpret and orchestrate, so nobody has to hand-write the logic that turns a human request into specific API calls.

1. A HubSpot Super Admin approves the connector and sets its data permissions (which objects it can read and write).
2. In Claude, go to Settings → Connectors → Browse → HubSpot → Connect, log into your HubSpot account, and choose permissions.
3. Toggle HubSpot on in a chat. Claude can now read contacts, deals, campaigns, marketing emails, and engagement history, and create or update records.

Our recommendation: set “Write tools” to Needs Approval so every write has a confirmation step.

Salesforce: CLI-based, through Salesforce’s own MCP server

Salesforce publishes an MCP server (@salesforce/mcp) built on the Salesforce CLI (sf) toolchain. We connect Claude to Salesforce through the command line, which means Claude can query your org, inspect metadata, and build or modify things directly, all from a conversation.

The rest of your stack

The same MCP pattern extends to almost anything with an MCP server: Google Drive and Sheets for supplementary data, Slack for distributing reports, email tools for send automation, and Chili Piper (which has its own MCP). Claude Code can hold multiple MCP connections at once, so a single agent session can span HubSpot, Salesforce, your data warehouse, and Slack in one workflow.

What you can do once it’s connected

1. Reporting and dashboards

Claude can query HubSpot or Salesforce directly (contacts, deals, campaigns, engagement history) and build a report or chart from a single prompt, instead of an analyst pulling CRM exports into a spreadsheet.

• Recurring reports such as weekly pipeline review, MQL-to-SQL conversion, and campaign ROI by channel can become repeatable prompts or scheduled workflows. The “build me this dashboard every Monday” task that used to eat an analyst’s morning goes away.
• Cross-system analysis becomes conversational. Claude can reason across both platforms at once, for example: “show me deals in Salesforce that originated from HubSpot campaigns with below-average email engagement.” Neither tool does that well natively.
• Dashboard building and refinement. In one recent example, we pointed Claude at an existing Salesforce dashboard via the Salesforce CLI and asked it to rename the dashboard to something more meaningful based on the reports inside it, and to recommend additional report widgets worth adding.

2. Investigating lead routing issues and lost leads

This is one of our favorite use cases. No more manually digging through your Salesforce instance and combing a lead’s activity log to work out where and why it got lost. With a simple prompt, Claude can tell you.

A real example: this very morning, a salesperson’s last name changed after she got married, and it broke lead routing. A lead was not routed correctly. Claude identified the cause within 20 seconds.

The same approach works for broader investigations into lead sources and for finding what is quietly broken in your tech stack. Here is a sample prompt we use:

“Use Salesforce CLI and find leads that are either in the missing channel source or missing lead source widgets, and try and determine what their lead source and channel source will be. Confirm with me what you think it is first before you push it live into Salesforce.”

Notice the last sentence. We always have Claude confirm before it writes anything back.

3. Building workflows in Salesforce

Because Claude works through the CLI, it can help you build and modify Salesforce workflows, not just report on them. This is where you start to feel the shift from “AI that answers questions” to “AI that does ops work.”

4. Campaign operations

Drafting marketing emails, building campaign asset lists, and updating campaign metadata are necessary but repetitive tasks. Claude can draft the copy and, through the connector, create or update the actual marketing email or campaign record in HubSpot. A human still reviews before anything goes live.

Lead and contact list building, deal-stage updates, task and note logging, and routine CRM hygiene (deduping, updating lifecycle stages) can all be delegated with human approval checkpoints.

The net effect: junior analysts shift from “data plumber” to “reviewer and strategist.” They check Claude’s draft dashboard or campaign build rather than assembling it from scratch.

Bonus: Chili Piper for faster meetings

A few of our clients have seen a big benefit from onboarding Chili Piper. Prospects who land on the website can book a meeting immediately, and it is routed to the right sales rep. That removes the back-and-forth over scheduling demos and intro calls.

After just one week of implementation, we have seen more meetings booked, fewer no-shows, faster sales cycles, and faster closed-won deals. Chili Piper also has an MCP, so it can plug into the same Claude workflows, and we have even built Salesforce dashboards that track Chili Piper responses through to pipeline.

Which Claude model should you use?

Claude comes in several versions, and the model picker describes each one well:

• Fable 5.1: for your toughest challenges. The most capable model, but it draws down your usage much faster than Opus.
• Opus 5.5: for complex tasks. Our day-to-day pick for most CRM investigations and builds.
• Sonnet 5.5: the most efficient for everyday tasks.
• Haiku 4.5: the fastest, for quick answers.

We typically stick with one workhorse model for ops work because it handles the task efficiently without burning through our usage. Save the heaviest model for the genuinely hard problems.

[Screenshot: the Claude model picker]

What does it cost?

Less than most teams expect. Typical plan options:

Plan	Approximate cost
Claude Max (20x)	$200/month
Team Standard	~$25/seat/month
Team Premium (includes Claude Code)	~$125/seat/month

If you are using the API with Opus, pricing is $5 / $25 per million tokens (input / output). In our experience, we don’t exceed $200 per month. Keep in mind you can hit usage limits, especially on the heaviest model, so match the model to the task.

One important caveat: keep humans in the loop

It is very important to have the right skills on your team so someone ensures data is being stamped correctly and coming into your systems correctly, and to always QA-check your reports. Claude is getting better at this every month: it is increasingly good at choosing the most likely fields to build reports, and at generating new campaigns for campaign operations.

But these are mission-critical systems that impact your company’s revenue and bottom line. Members of your marketing operations team still need to QA the output and monitor how Claude generates and creates workflows and agents in your systems. Treat Claude like a very fast, very capable junior teammate: great at the first draft, and still in need of a reviewer.

Want help setting up an AI-driven marketing ops stack? Get in touch at Mopsify.com.