Claude Cowork and your CRM: what's possible when your AI agent can browse

We’ve been working with a cutting edge client who loves collecting all data possible. They had an issue; they wanted to automate reporting on their growing LinkedIn pages, but the analytics sits behind a login. There is no export button for post engagement data, and the platform’s native analytics tools are frustrating if you want to track performance across multiple posts over time.

Another data source they use is a third party contact enrichment tool, to solve a related problem: they visit profiles, read company data, and push it into their CRM. Useful, but these enrichment tools often charge per contact or per month, and they shape their output to a generic data model, not yours. A client conversation this week pointed to a different approach, built around Claude Cowork and a CRM MCP connection.

Contents

• What Claude Cowork actually is
• Getting data out of places that don’t have APIs
• Where your CRM’s MCP comes in
• Building your own enrichment tool
• The loop that becomes possible
• Who this actually suits right now
• Why read-only first is the right call
• TL;DR

What Claude Cowork actually is

Claude Cowork is a browser agent that runs inside Claude and can navigate the web the way a person would. It can open a page, log in with saved credentials, read what is on screen, interact with elements, and hand the results back to your Claude project. Unlike a Zapier trigger or a native integration, it does not need a structured API endpoint to connect to. If a person can see the data in a browser, Cowork can reach it.

It’s worth noting that this is still early-stage software. The capability is genuinely powerful but the setup is not yet click-and-go, which matters for assessing whether it suits you right now.

Getting data out of places that don’t have APIs

As with many of our clients, this five-person software consultancy we engaged with on a CRM migration project uses Claude as a sales thinking partner. They had an immediate data problem: LinkedIn post analytics sit behind a login, and there is no bulk export for engagement data. They wanted to track performance across posts over time and share a weekly summary with the Directors.

Here’s their solution with Claude Cowork: save the analytics URL for each post in a local file that Claude has access to. On demand, they ask Claude to fetch the latest numbers. Cowork opens each saved URL, logs in, reads the engagement data on screen, writes the results to a CSV, and produces a formatted report. Manual trigger, no schedule, no third-party export tool required.

It is a specific workflow, but it points toward something broader. There is of course scope to automate this so it runs on a regular cadence. One quirk of Cowork at the moment that our client mentioned is that the laptop or machine Claude is running on has to be awake for Cowork to run its tasks, so this is worth keeping in mind.

Where your CRM’s MCP comes in

An MCP (Model Context Protocol) is a structured connection that lets Claude read from and write to another system. Think of it as a purpose-built channel between your AI agent and a specific tool. Several CRMs now have MCPs available or in active development, which means Claude can query your contact data directly without you copying anything into the conversation manually.

Here is where the main platforms stand:

• Capsule CRM has an MCP currently in beta and read-only, meaning Claude can find contacts, pull open deals, and check field values. Write access is in development.
• HubSpot has an official MCP server available. It gives Claude read and write access to contacts, companies, deals, and notes.
• Pipedrive has an MCP-compatible integration, allowing Claude to query deals and contacts from your pipeline.
• Zoho CRM has MCP support via its API layer, covering contacts, leads, and activities.
• Attio has a well-documented API and community MCP support, which suits Claude integrations well given Attio’s flexible data model.

Read access on its own is already useful. Ask Claude a question about your pipeline or your contacts and get an answer drawn from live CRM data rather than an export you ran last Tuesday. Combine read access with Cowork and the picture changes considerably.

If you are weighing up which CRM to use alongside this kind of workflow, our guide to whether Capsule CRM is the right fit covers what it suits best. For a broader comparison, our HubSpot alternatives post covers where the other platforms sit.

Building your own enrichment tool

This is what the consultancy’s next focus is, and the plan is straightforward: point Cowork at a list of LinkedIn profiles, have it read role, company size, and recent activity from each one, and use the MCP to write that data back into the CRM as structured contact fields or notes.

This means no per-contact fee. No subscription to a third-party enrichment service with a fixed monthly bill. No data being processed by a platform whose data model you did not choose. The consultancy had been about to pay around £20 a month per user for a dedicated LinkedIn-to-CRM enrichment tool. This combination makes that unnecessary, and saves them money on their software subscriptions.

The difference from a Zapier workflow is worth making explicit. Zapier makes it easy to connect apps that have APIs. It can move data between systems when both ends have a structured integration available. Cowork handles the gap that Zapier cannot reach: pages that require a login, present data visually, but offer no API. The MCP then provides the structured write target that turns scraped data into CRM records your whole team can see and act on.

The loop that becomes possible

Take the workflow a few steps further and a complete content-to-CRM loop becomes plausible. You publish a post on LinkedIn. Cowork fetches the engagement data and logs it. People who engage get visited by Cowork: company, role, recent activity read from their profiles. The CRM MCP writes that enrichment back into your contact records. From there, a nurture sequence picks up the contact and starts building the relationship over time, whether that is a Transpond sequence in Capsule, a HubSpot workflow, or a Pipedrive activity cadence.

None of these steps are fully automated against each other yet. Write capability on most MCPs is either limited or still arriving. But the architecture is coherent, and understanding it now means you can build toward it rather than retrofitting everything once the tooling catches up.

For the downstream end of that loop, our B2B email nurture guide covers what to send once the contact is in your CRM and the timing that works for longer sales cycles.

Who this actually suits right now

The consultancy we spoke with is comfortable with Docker, coding, and command-line tooling. That context matters, as they are definitely towards the technical end of our clients! The pattern they have built is genuinely useful and generalisable. However their specific execution is not yet accessible to everyone, especially those who are less technical.

To run Claude Cowork alongside a CRM MCP today, you need a Claude account with Cowork access, some comfort configuring local tools and files, and a CRM that has an MCP available. This is not an easy-to-install plug-in, in fact it’s closer to a lightweight technical project. The experience will improve as the tooling matures, but right now it sits in the ‘technically confident’ segment of the small business market.

TL;DR

• Claude Cowork is a browser agent that can navigate logged-in pages without needing an API. It reaches data that traditional integrations cannot.
• A five-person consultancy uses it to pull LinkedIn post analytics on demand and produce a formatted report, replacing a manual process with no third-party tool.
• CRM MCPs let Claude query your CRM directly. HubSpot has the most complete MCP available now. Capsule is in beta (read-only). Pipedrive, Zoho, and Attio all have MCP-compatible integrations at various stages of maturity.
• Combining Cowork with a CRM MCP creates a contact enrichment workflow that replaces paid LinkedIn-to-CRM tools at no per-contact cost.
• The full loop, from post to engagement data to enriched CRM contacts to nurture sequence, is architecturally plausible now. Write capability across the platforms is the remaining piece.
• This is currently best suited to technically comfortable users. The tooling is early-stage and not yet a no-code experience.