In a snapshot: Capsule has built a Model Context Protocol server, letting AI tools like Claude read your CRM data directly. It is read only, it runs on your own machine, and it is in private beta for Growth plan customers. Small feature, big signal.
What Capsule announced
The first mention appeared on 15 April 2026, tucked away in Capsule’s developer changelog. A local MCP server, in private beta, giving MCP compatible AI tools such as Claude Desktop and Cursor read only access to your parties, opportunities, tasks and projects through the Capsule API.
Capsule gave it a proper launch on 26 June with a blog post and an early access sign up. The pitch is simple: open Claude, type a question like “which deals are closing this month but have not had activity in the past fortnight”, and get an answer pulled from live CRM data. No report builder, no filters.
MCP in plain English
MCP stands for Model Context Protocol. It is an open standard, originally published by Anthropic, that gives AI tools a common way to connect to the software you already use.
Before MCP, every AI tool needed a custom built connection to every application. MCP replaces that with one shared plug. Build the server once and any compliant AI tool can talk to your data. Think USB rather than bespoke integration project.
Why this matters for smaller businesses
Reporting is the part of CRM most small teams never get round to. You know the data is in there. Building the report to get it out is a job for a Tuesday afternoon that never arrives.
Asking a question in a chat window removes that step. For an owner managed business with no operations person and no analyst, that is a genuine unlock. The examples Capsule gives are the things people actually want to know: which deals are stalling, which contacts have gone quiet, which projects have no owner, who logged the fewest activities last month.
Read the small print
A few things to be clear about before you get too excited.
It is read only.
The server can answer questions. It cannot update a record, log an activity or move a deal. That is a sensible first step for a vendor and a real limit for you. Anyone hoping to say “log that call and set a follow up for Thursday” will have to wait.
It runs locally.
The server is a Docker container on your own machine, and data passes between your AI tool and the Capsule API without going through an external server. Good for privacy, awkward for anyone who has never installed Docker.
It is gated.
Private beta, access by request, Growth plan or above. Free during the beta, although your AI tool’s own subscription still applies. You request access from Integrations inside your Capsule account.
Two clients only.
Claude and Cursor today. No ChatGPT.
Governance still applies.
Your customer data becomes readable by a third party AI tool. Check what that vendor retains, whether it trains on your inputs, and whether your privacy notice covers it. Read only reduces the risk, it does not remove it. Remember too that the AI sees whatever the connected user can see, so think carefully about who gets it.
The bigger picture
Capsule is a small vendor moving early, and that is the interesting part. The larger platforms are heading the same way. Within a year I expect an MCP endpoint to sit on CRM requirement lists next to API access and webhooks.
That changes the selection conversation. For the past two years the question has been “what AI does this CRM have built in”, and the answer has usually been an email drafting assistant and some deal scoring. An MCP server flips it round. You bring your own AI, and the CRM becomes a data source you can interrogate with whichever tool you already pay for.
For an SME already paying for Claude or similar, that is better value and far more flexible than waiting for a vendor to ship features you never asked for.
What to do next
If you are on Capsule and on a Growth plan or above, request beta access and try it on the questions you normally avoid building reports for. Give it a fortnight before you decide whether it changes anything.
If you are choosing a CRM this year, put MCP support on your requirements list. Ask whether it exists, whether it is read only or read and write, which AI clients it supports, and where the data goes.
And if you are on another platform, go and check the changelog. These things tend to arrive quietly, in a developer note, months before anyone writes a blog post about them.
Want an independent view?
We are vendor agnostic. We hold partnerships across the major CRM platforms, which means we have no reason to push you towards one or away from another. Book a free no obligation consultation if you are weighing up whether your current CRM is keeping pace, or you are choosing one this year and want a straight answer rather than a sales pitch, we are happy to talk it through.
