Operations are included
After handover you get access to a portal where you can see the state of your integration at any time. Not a PDF report once a month.
Model Context Protocol
AI connected to your business systems.
We design, install and maintain MCP servers — the layer that lets AI work safely with your CRM, email, documents, helpdesk or internal API. You decide what it may do. Everything else stays closed.
- Ready-made connectors
- for common tools
- Custom MCP server
- for internal systems
- Operations and oversight
- after handover
AI assistant, agent, team chat
asks in plain language
MCP layer
allowed operations, role-based scope, call log
CRM · email · helpdesk · documents · internal API
your data stays with you
In short
What we do
We build the layer through which AI works with your systems — and then we run it.
- We pick and deploy an existing MCP server, or write one for your in-house system.
- We connect AI to your CRM, e-mail, documents and internal API — within the scope you define.
- We keep it running: measuring, watching and fixing when something breaks.
What is different
Plenty of people can build an MCP server. This is what tends to be different here.
If we do not know, we do not claim
The portal shows measured state. Where a measurement is missing we say so, instead of filling in an estimate you would needlessly trust.
Permissions are verified
The server itself reports which tools it actually offers. We compare that with the approved list — an extra tool is handled as an incident. And we never store your credentials at all.
You talk to the builder
Design, implementation and operation in one place. No handover between sales and delivery, and no arguing about who promised what.
Why MCP
AI without access to your systems only knows half the job.
It can write and summarise. What it doesn't know is your customers, your orders, or what was agreed by email yesterday. The MCP layer gives it that context — in the scope you define. This isn't about replacing people: it's about making sure that shuffling data between systems by hand stops eating the time that belongs to customers, exceptions and decisions.
Today
Extra work nobody asked for
- People retype data between tools and keep the same information in three places.
- AI answers in generalities because it can't see the current state of a customer or an order.
- Information is scattered across email, CRM and documents.
- The internal system works fine, but nobody knows how to connect it to modern AI tools.
With an MCP layer
AI that works with real data
- look up information in the CRM
- prepare material from email and documents
- work with an internal API
- automate repetitive tasks
- stay within clearly defined permissions
Every operation is named and approved in advance. If it isn't on the list, the AI cannot do it.
Scenarios
What this means on an ordinary working day.
Six situations we're asked about most often. The exact permissions and allowed actions are set per company in each of them.
Sales
Summary of the customer conversation
The AI goes through CRM history and the email thread and writes up what matters — what was agreed, what is pending and where it stalled.
- CRM
Sales
Current order status without the hunt
A salesperson asks in plain language and gets an order status assembled from several systems instead of opening them one by one.
- CRM
- internal API
- documents
Support
Case history and a draft reply
The AI finds earlier tickets and the related documentation and drafts a reply. Sending stays with a human, if that's how you set it up.
- helpdesk
- knowledge base
Sales
Quote material from internal data
Price list, earlier quotes for a similar customer, current stock or capacity — the AI assembles a draft that only needs checking.
- price list
- CRM
- database
Management
Answers based on current operational data
Management asks against live data instead of a month-old export. The scope is bounded by role, not by an agreement about what nobody will ask.
- database
- reporting
IT
Connecting your own application without exposing an API
The internal system is reachable only through specific named operations. The AI gets exactly those — not the whole database.
- internal application
- database
What we deliver
From picking a connector to watching over production.
Most companies don't need everything at once. We start where the payoff is fastest and add the rest when it makes sense.
Installing existing MCP servers
01MCP servers already exist for common tools. We pick the right ones, configure them and deploy them safely into your environment.
- connector selection based on the tools you actually use
- configuration, credentials and permission scope
- deployment to your server or an environment of your choice
- verification that the AI handles your data the way you expect
Custom MCP server development
02For an internal system, a legacy application or a specific process we write a connector built exactly for it.
- analysis of the data and operations the AI needs to handle
- design of the tools and their inputs and outputs
- integration with your API, database or application layer
- a test environment kept separate from production
Business tool integration
03We connect AI to what you actually work in — CRM, email, helpdesk, documents, databases and internal applications.
- CRM and sales records
- email, shared mailboxes and calendars
- helpdesk, ticketing and knowledge bases
- documents, databases and internal web applications
Maintenance and support
04An integration isn't a one-off install. We look after operations, updates and further development — and you watch the state of your servers in your own portal, not in a monthly report.
- availability and error monitoring
- updates to the servers and their dependencies
- permission reviews when your team or systems change
- extending the setup with more tools as priorities shift
Who it's for
Two different briefs, one integration layer.
For companies without in-house development
Tell us what the AI should do. We'll handle the technical side.
You don't need to understand protocols or server operations. Just describe what slows you down today and which systems you use. The rest — design, installation, permissions and ongoing maintenance — is on us.
- a plain-language proposal without jargon
- installation and hosting on our side or yours
- one partner for design, delivery and support
For IT teams
An integration layer that fits your environment and your rules.
We respect how you've set up authentication, environment separation and approvals. Code and configuration are handed over with documentation, so you can take operations in-house whenever you want.
- service accounts, token rotation, separate environments
- tool-call logging and a defined set of operations
- deployment into your infrastructure, documentation for handover
Case studies
What already runs in production.
Four deployments we look after. Described in general terms — client names only with their consent.
Healthcare
Booking system of a dental practice
An MCP server on top of the practice's booking system. The AI sees free slots, creates and moves appointments and watches the practice's capacity. There's also a bridge into ChatGPT over its own OAuth, so staff can use the assistant they already know.
What changed
Appointments are created during the conversation with the patient. Nobody retypes them from a phone call into the system.
What the AI may do
- read the calendar
- create and move appointments
- nurse role, not admin
- medical records out of reach
Coworking and training
Running a coworking and training centre
An MCP layer over roughly fifty internal API paths: room occupancy, course dates and capacity, memberships and billing material. Its own service account with its own scope, separate from staff accounts.
What changed
A question about a free room or a course date is answered right in SMS or chat — without opening the admin.
What the AI may do
- occupancy and bookings
- course schedule
- billing material
- service account with its own scope
IT services and hosting
Automated support across 3,500 clients
The AI answers clients' email enquiries. It draws on a knowledge base that automation keeps up to date, and opens and updates the related cases in the ticketing system. What's left for support is what genuinely needs a person.
What changed
Routine questions get answered without waiting for a free person, and the answer leaves a trace in a ticket.
What the AI may do
- incoming client enquiries
- knowledge base
- writing to the ticketing system
- every call logged
Sales and marketing
Automated sales outreach
The largest of the deployments, 36 tools: the AI works with leads, reads communication history, writes personal outreach and quotes, triages incoming replies and logs activity. Write operations were enabled gradually, one at a time.
What changed
Outreach runs on its own and the salesperson steps in where it counts — with a prospect who replied.
What the AI may do
- leads and history
- personal outreach and quotes
- triaging replies
- logging activity
How we work together
Four steps from the first call to a running solution.
- 01
Initial consultation
We go through the systems you use, what the AI should do and what your security rules are. We work out where an integration makes sense and where it doesn't.
Output: list of systems and priorities
- 02
Solution design
We pick ready-made connectors where they suffice and design a custom MCP server where they don't. Permission scope is part of the design.
Output: solution design and budget
- 03
Implementation and testing
We connect the systems, set up access and verify everything in a test environment against real scenarios first.
Output: scenarios verified in test
- 04
Handover and maintenance
We hand over documentation and train your users. Then we stay with the running system — monitoring, updates and further development.
Output: documentation and monitoring
We usually start with the single process that costs you the most time today. It proves itself in production — and only then do we decide about the rest.
Pricing
Three models, all derived from one hourly rate.
The basis is an hourly rate of CZK 700. Project work and the retainer both come from it, just packaged differently. No surprise items, no “we'll agree on that later”.
01 — Consulting
Hourly
fromCZK 700/ hour
Technical consultation, process audit, review of an existing integration, solving a specific problem.
- minimum billing unit 30 minutes
- timesheet of hours worked
- first half hour free
02 — Project
Fixed price
estimateCZK 700× hour
Deploying an MCP server, building a custom connector, integrating a system. A fixed budget estimated from the hourly rate.
- price agreed up front
- milestones and interim deliverables
- 30 days of warranty support after handover
03 — Operations
Monthly retainer
fromCZK 3,500/ month
Looking after a solution that already runs: monitoring, updates, permission reviews and small improvements.
- an agreed block of hours (e.g. 5 h/month)
- response within 4 hours in business hours
- unused hours carry over
Not VAT registered, payment terms 14 days. The first half hour is free — so we both know whether it's worth continuing.
Security and control
Access scope is a design decision, not a side effect.
We don't promise absolute security — that wouldn't be honest. We promise it will be clear what the AI may do, where that is written down and how to change it.
We always review the design against your internal rules. If something conflicts with your security policy, we say so before it gets built.
Least necessary privilege
Each tool gets only the operations and data it genuinely needs for its job. The default is “no access”, not “access to everything”.
Role-separated access
One scope for sales, another for support, another for management. Roles follow the rules your company already has.
Handling credentials
Tokens and passwords live on the server — not with the user and not in the language model. They're stored outside the repository and can be revoked and rotated at any time.
Auditability
Tool calls are logged. How far a change can be traced inside the target system depends on what that system itself allows — we agree on that up front.
Separate environments
Test and production have their own credentials. Scenarios are verified in test before they touch real data.
Monitoring and updates
We watch availability and errors, update the servers and their dependencies, and review permissions when your team changes.
After go-live
Someone is accountable for running the integration — and you can see it.
- availability and latency measured every minute, from outside
- versioned configuration: who changed which permissions, and when
- your own access to the portal, not a monthly report
FAQ
What companies ask most often.
What is an MCP server?
MCP (Model Context Protocol) is an open standard for connecting AI tools to data and systems. An MCP server is a small service that sits between the AI and your system and offers clearly named operations — for example “find customer” or “list open tickets”. The AI can only use the operations the server exposes to it.
Do we have to change our CRM or email system?
No. The MCP layer connects to what you already use, through an existing API or database. If your system has no interface, we look at it case by case — sometimes the connection can be built another way, sometimes the honest answer is that it doesn't make sense.
Can the AI make changes in our systems?
It can, if you allow it. We separate read operations from write operations and enable writes deliberately, one at a time. Sensitive actions usually get human confirmation — the AI prepares a draft and a person approves sending or saving it.
Does this mean AI takes over our people's work?
Not in the sense of replacing them. What the MCP layer removes are the steps that keep repeating — looking up records, retyping data between systems, assembling the same background material again. Decisions, exceptions and dealing with the customer stay with people, who get more room for them and less switching between tools. Write operations are enabled deliberately, and sensitive actions are commonly set up to require human confirmation.
How are access and permissions handled?
The server has its own service account with a role-derived scope, not full administrator access. Credentials live on the server, not with the user. The scope can be narrowed, widened or revoked at any time.
What does it cost?
The basis is an hourly rate of CZK 700, from which both the fixed project price and the monthly retainer are derived — the full pricing is a little further up this page. The work is split into three phases: Discovery (paid thinking: naming the real problem, proposing a solution and a clear yes/no decision), Sprint (the build itself, as a fixed-price package or a block of hours) and Care (long-term care of a running solution). A Sprint never starts without Discovery — without it, any price would be a blind guess.
How long does it take?
Discovery is a matter of days and ends with a concrete recommendation. The Sprint for a first integration is usually weeks, not months — which is why we suggest starting with one process and expanding based on what production teaches you. You get the schedule as part of the proposal, not after signing.
Can we start with a single process?
We recommend it. Pick the one activity that costs you the most time today, we'll connect it and prove it in production. Only then do you decide about the next steps — with real experience instead of guesswork.
Do you also provide long-term maintenance?
Yes, and we consider it part of the solution. An integration ages along with the systems it connects: APIs change, tokens expire, new people arrive. You can keep maintenance with us, or take the solution in-house with documentation.
Can you build an MCP server for our own system?
Yes, that's a routine part of the work. We need a description of the system and access to its interface or database. Together we pick the operations the AI should handle and implement those — the rest of the system stays out of reach.
Why work with us
An integration lives or dies by whoever runs it.
Building the connection is the easy part. Keeping it running is harder — when an API changes, a token expires or a new person joins the team.
- Over 20 years in IT — from servers and networks to running applications that have to stay up at weekends too.
- I work with Linux, automation and AI integrations every day. I build and run MCP servers, not just read about them.
- You deal directly with the person who designs and builds the solution. No anonymous agency, no handoffs between departments.
- Design, implementation and long-term maintenance in one place — including responsibility for keeping it running.
- We can work in Czech, English or German.
Contact
Martin Šabata
IT, AI & Automation Freelancer
- Phone
- +420 774 509 038
- Address
- Velešovice 33
683 01 Velešovice - Company ID
- 73584088 · Not VAT registered
Enquiry
Tell us about the system you want to connect to AI.
A few sentences are enough. I'll come back with a proposed next step, or ask about what matters. The consultation is free and non-binding.
Pick a call slot
Prefer to talk? Pick a free slot — the first 30 minutes are on us.