{"slug": "your-crm-is-going-headless", "title": "Your CRM is going headless", "summary": "HubSpot's remote MCP server became generally available on April 13, 2026, letting any MCP-compatible AI assistant read and write contacts, companies, deals and tickets, a move the company frames as embracing the shift toward \"headless\" CRMs. HubSpot chief product and technology officer Duncan Lennox wrote that \"Connectors for Claude, ChatGPT, Gemini, and Copilot are delivering real value to customers\" and set the standard that \"anything you can do inside HubSpot, you should be able to do through an API.\" The article argues that while agents handle structured fields well, the context reps rely on to close deals — what was promised on a call, why a deal stalled — still lives outside the CRM in email, Slack and calls.", "body_md": "# Your CRM is going headless\n\nYesterday I asked [why people use AI chat for everything](https://www.thisandthat.chat/blog/why-people-use-ai-chat-for-everything/). The clearest case of it I see right now is the CRM. A growing number of the people we talk to still pay for HubSpot or Pipedrive, yet rarely open it. The CRM holds the records, but they ask Claude or ChatGPT about their pipeline, their accounts and what changed last week. The CRM has gone headless: it’s the database, and the interface is somewhere else.\n\n## Why the legacy CRMs went headless\n\nI’ve started calling HubSpot and Pipedrive legacy CRMs, which may sound harsh for products that ship updates constantly. I mean something narrow. They were designed around people filling in records by hand, and agents arrived later. That’s why people use them headless.\n\nHubSpot has its own AI, Breeze, with agents that work inside the product. Few of the people we talk to take it very seriously. They already have a favorite assistant, and it can see their own email and documents as well as the CRM. Asking it about a deal is one more prompt in a window that’s already open.\n\n## HubSpot seems to agree\n\nHubSpot appears to understand this well. Its remote MCP server [became generally available on April 13, 2026](https://developers.hubspot.com/changelog/remote-hubspot-mcp-server-is-now-generally-available). MCP is the protocol AI tools use to connect to other software, and through it any compatible assistant can read contacts, companies, deals and tickets, and create or update them “through natural conversation.” In [HubSpot’s post on its vision for the agent era](https://www.hubspot.com/company-news/our-vision-for-building-an-open-ecosystem-for-the-agent-era), Duncan Lennox, its chief product and technology officer, writes that “Connectors for Claude, ChatGPT, Gemini, and Copilot are delivering real value to customers.” He also sets a standard: “anything you can do inside HubSpot, you should be able to do through an API.”\n\nI think it’s a sensible bet. If people are going to run their CRM from somewhere else, the CRM that agents find easiest to read and write stays the system of record. HubSpot is still building Breeze too, and the same post talks about making that work available wherever teams and agents operate.\n\n## What the record leaves out\n\nA headless CRM is good at what it has always held: names, stages, amounts, close dates and the activity log. Agents read those well, and now they can write them.\n\nThe knowledge a rep leans on to close a deal is mostly somewhere else. What was promised on the last call. Why the deal stalled in September. Who on the customer’s side actually decides, which is often someone other than the person with the title. Some of it has a field, like a closed-lost reason, but the rest has no obvious home in the record. It arrives in email, in Slack and on calls, from the customer and from colleagues. Some of it reaches a notes field if a rep is especially diligent, if a sales ops team keeps things in sync, or if a sales leader insists that people document things thoroughly, but that last one only works with consistent nagging.\n\nA former sales VP told me recently that it’s odd to keep any customer data outside the CRM. It may be odd, but it’s common. One person we spoke to ran HubSpot headless and still kept plenty of customer information outside it.\n\nBefore AI, reps updated CRM records by hand. When they did, they at least had a chance to add that context while they were there. When an agent answers from the CRM, it answers from what’s in the fields and in your own inbox, while the promise may sit in a colleague’s thread or a Slack message you weren’t part of.\n\n## So where does that context go?\n\nIf the CRM becomes a backend that agents read and write, something has to hold what doesn’t fit in the record. It could be the CRM itself. Attio, for one, already captures emails, calls and meetings onto the customer record as they happen. It could be the assistant’s memory, which usually holds what you told it rather than what your colleague heard on Tuesday. Call recorders and shared docs hold some of it too, though a recorder only hears the calls it joins and a doc only holds what someone writes up. Or it could be a layer alongside the CRM that reads the team’s conversations and keeps track of what they say about each customer, for whichever person or agent asks.\n\nA layer alongside the CRM is what we’re building at this+that. Our Company Brain reads your team’s email and the Slack channels you choose, and keeps a page for each customer: what was decided, who owns what, what’s due and what’s in the way, each fact citing the message it came from. The CRM stays the system of record for stages and amounts. The Company Brain holds the context around them, and Claude and ChatGPT can read it through our MCP server. We’re also planning apps and connectors for CRMs such as HubSpot, Pipedrive and Attio, so the same context can appear on the CRM record and, where the CRM allows it, be available to its own AI, like HubSpot’s Breeze.\n\nIf you run your CRM headless, where do the promises from calls and the name of the real decision-maker end up for your team? And if you’ve stopped opening your CRM, have you noticed anything that stopped reaching it?", "url": "https://wpnews.pro/news/your-crm-is-going-headless", "canonical_source": "https://www.thisandthat.chat/blog/your-crm-is-going-headless/", "published_at": "2026-11-20 10:14:59.211624+00:00", "updated_at": "2026-11-20 10:15:01.906599+00:00", "lang": "en", "topics": ["ai-agents", "agent-protocols", "ai-products", "ai-tools"], "entities": ["HubSpot", "Pipedrive", "Breeze", "Duncan Lennox", "Claude", "ChatGPT", "Gemini", "Copilot"], "also_reported_by": [], "takeaway": "HubSpot's remote MCP server, generally available since April 13, 2026, lets any MCP-compatible assistant read and write contacts, companies, deals and tickets, so teams can run the CRM from Claude, ChatGPT, Gemini or Copilot instead of HubSpot's own interface.", "recommendation": null, "alternates": {"html": "https://wpnews.pro/news/your-crm-is-going-headless", "markdown": "https://wpnews.pro/news/your-crm-is-going-headless.md", "text": "https://wpnews.pro/news/your-crm-is-going-headless.txt", "jsonld": "https://wpnews.pro/news/your-crm-is-going-headless.jsonld"}}