Every headline this year says AI is changing how companies operate. Then Monday comes around, someone updates the Scorecard by hand, and a Rock owner gets asked for the third week running whether that project moved.
It has to do with where your operating data lives. Your Rocks, your measurables, your Issues list — all of it sits inside an application that an AI assistant can’t see.
MCP (Model Context Protocol) is an open standard that connects an AI assistant to the systems where your data lives. For a company running EOS, an MCP connector lets Claude or ChatGPT read and update your Rocks, Scorecard, To-Dos and Issues directly, using your existing permissions.
We shipped one earlier this year. Most of the operators who beta-tested it for us had never heard the term before we emailed them 🤔. Below is the explanation we wish had existed then — what MCP is, what it changes for an EOS company, and the parts vendors leave out.
What MCP actually is MCP is an open-source standard, introduced by Anthropic in November 2024 and now supported across Claude, ChatGPT and a long list of other tools. The protocol’s own documentation offers the analogy that stuck: think of MCP as a USB-C port for AI applications — one standard shape, so anything plugs into anything.
The problem it was built to solve is worth quoting directly. Anthropic’s announcement described even the most capable models as “constrained by their isolation from data — trapped behind information silos and legacy systems.”
That describes an EOS company accurately. Your Scorecard is one of those silos, along with your V/TO, your Issues list, and every Rock description someone wrote three quarters ago that nobody has opened since.
The commercial point is worth understanding too. MCP is an open standard, so any vendor can build one and any assistant that supports the protocol can connect to it. Nothing about it ties you to a single tool.
“No. To be honest I didn’t know what an MCP was when you guys emailed me.” — Shaz Khan, Integrator, the hospitality group behind Frank & Andrea, Tono and Slicecraft
Shaz was one of our first beta testers, and not knowing the acronym slowed him down for about a day. He went on to build a dashboard that pulls his EOS data together with weather forecasts to plan staffing across a dozen-plus locations.
What MCP means for a business operating system A business operating system turns out to be an unusually good fit for this, and the reason is structure.
Most company data is a mess for an AI to reason about. Notes in a doc, numbers in a spreadsheet, decisions in a thread. EOS data is different — it arrived pre-organized. Rocks have owners and due dates, measurables have targets and a weekly cadence, Issues have a status, and To-Dos come with a seven-day clock.
That structure is what makes the answers reliable. When you ask which Rocks are at risk, the assistant doesn’t have to work out what “at risk” means — your operating system defined it two quarters ago.
The shift, in practice, is that you stop navigating to an answer and start asking for it.
Ours is called the Strety MCP , and it reads and updates eight things: To-Dos, Issues, Headlines, Scorecard measurables, Rocks, Docs, Agendas and Projects. Setup takes four steps in your AI tool’s connector settings and about two minutes. We’ve written up nine things operators actually do with it if you want the specifics.
How this differs from your existing integrations You already have integrations. Your PSA talks to your accounting system, your calendar talks to your meeting agendas, and none of that is new.
An MCP connector works differently:
Traditional integration MCP connector What moves Records, on a schedule Nothing — the assistant reads live Who defines the questions Whoever built the integration You, when you ask Setup An admin configures it for the company Each person connects their own When you need something new Someone builds it You ask differently
The practical difference lands in that second row. An integration answers questions somebody anticipated when they built it. An MCP connector answers the question you have on a Thursday afternoon, including the weird ones. Try this one: which Rock owners have gone two weeks without an update, and what did they commit to in the L10 before that?
Both matter. The integrations that keep your data moving between systems still do their job, and this sits alongside them.
What it looks like on a Tuesday Two examples from our own week.
The morning read. Before you open anything, you ask what’s red on the Scorecard and which Rocks are at risk. You get a paragraph instead of six clicks and a tab you forgot you had open — games of inches.
Capture from the car. You’ve just left a client and remembered a commitment. You dictate it, and the To-Do lands in the right place with the right owner, before you’ve reached the highway.
Neither is dramatic. Both remove the friction that stops people from keeping their EOS data current, which is the failure mode in most implementations we see. Our co-founder now batch-audits Rock descriptions by voice, checking whether they’re specific enough to measure — nobody put that on a roadmap, he just started doing it.
If you want more, four operators described their setups in detail earlier this year.
What about security and permissions This is the question every operator asks, and skepticism here is the correct instinct.
McKinsey’s State of AI trust survey from March 2026 found that nearly two-thirds of respondents cite security and risk concerns as the top barrier to scaling agentic AI, with 74% flagging inaccuracy and 72% cybersecurity.
Time for some marketing here 🙂 Ours works like this.
You sign in to Strety to authorize the connection, and the connector inherits the permissions you already have. If you can’t see a team’s Scorecard in the app, the assistant can’t see it either. Setup is per person, so one operator can connect and try it without a company-wide decision or an admin project. And nothing gets copied — the assistant reads your live data through the connection rather than working from a duplicate sitting somewhere else.
Those three properties are worth confirming with whichever vendor you’re using.
What MCP can’t do It can’t fix an implementation that isn’t running. If your Scorecard is three weeks stale, the assistant will confidently tell you about three-week-old numbers.It can’t replace your L10. It takes the prep and the note-taking off your plate. Your team still has to sit down and solve issues together.It can’t reach data that isn’t in your operating system. If half your numbers live in one person’s spreadsheet, they remain invisible.It can’t make someone own their number. An owner who only knows their measurable because they asked an assistant this morning still doesn’t own it.The fourth one is the one we’d underline. A connector removes clicks; it doesn’t move accountability off a person.
What to ask your BOS vendor If you’re running on something other than Strety, ask these six — they’re fair questions to ask us too.
Does it read and write, or only read? Read-only is useful. Read-and-write is what removes the double-entry work.Which objects does it cover? A connector that reaches Rocks but not Issues will frustrate you within a week.Does it use my existing permissions? The answer should be yes, without hedging.Can one person connect without a company-wide rollout? This determines whether you can actually test it.Which assistants does it support? It should work with the one your team already pays for.Does it cost extra? Worth knowing before you build a habit around it.If a vendor can’t answer four of those six clearly, they shipped an announcement rather than a connector.
Frequently asked questions What does MCP stand for? MCP stands for Model Context Protocol, an open-source standard for connecting AI assistants to the systems where your data lives. It was introduced by Anthropic in November 2024 and is now supported across Claude, ChatGPT and other AI tools.
What does MCP mean for a company running EOS? It means you can ask an AI assistant about your Rocks, Scorecard, To-Dos and Issues in plain language, and have it update them, without opening your EOS software.
How is an MCP different from a regular integration? A regular integration syncs data between two apps according to rules someone set up in advance. An MCP gives an AI assistant live access to your data, so it can answer whatever you ask — including questions nobody anticipated.
Is it safe to connect an AI assistant to my company data? A well-built MCP connector uses your existing sign-in and respects the permissions you already have, so it can only reach what you could reach yourself. Ask any vendor to confirm that before you connect.
Do I need a developer to set this up? No. Connecting the Strety MCP takes four steps in your AI tool’s connector settings and about two minutes.
Can the AI change my EOS data, or only read it? Both. The Strety MCP reads and updates To-Dos, Issues, Headlines, Scorecard measurables, Rocks, Docs, Agendas and Projects.
Does everyone on my team have to use it? No. Setup is per person, so one operator can connect and try it without a company-wide rollout.
Does this replace our Level 10 meeting? No. It removes the prep and the note-taking around the meeting. The meeting is still where your team solves issues together.
Where to start MCP connects an AI assistant to the data you already keep. For companies running EOS it works unusually well, because the framework did the organizing work years ago: every Rock already has an owner, every measurable already has a target.
The useful next step is small. Connect it, then ask one question you’d normally open the app to answer. Two minutes, and you’ll know within one whether it changes your week.
The Strety MCP is included with every plan at no extra cost. If you’re not with us yet, you can try Strety free for 30 days and connect it on day one 🙂. And if you’re brand new to all of this, the complete guide to Strety is a better place to start than this post was.
