Skip to content
Home For whom What I offer Platforms & CMS Lab Writing About Contact
NL EN
The Connector · an MCP server on your own system

“I am losing my mind over all this retyping.”

Recognise that? Then your team works around its own software every day. Data from one screen to another, overviews assembled by hand, and an AI assistant nobody really uses because it knows nothing about your business. At Careerguide most of that work is gone, and below you can see how.

01

Every day I retype the same data from one screen into another.

02

I can no longer keep track across all those systems.

03

We have ChatGPT, but it knows nothing about our customers.

A pipe coupling with a red valve wheel on top.

01 · An MCP server on your own system

What is The Connector?

Between the AI assistant your team already has open and your own software sits a connector. It knows only the actions you agreed on, without a new screen and without migrating your data. At Careerguide this runs in production.

For the board

You decide which gates exist

A blue box with three toggle switches, one flipped and red.

Five to ten carefully designed actions. What is not in there the assistant cannot do, not even if someone asks cleverly.

For the employee

You work with your own permissions, nothing more

An access badge on a lanyard with a red clip.

Anyone who cannot see orders in the system cannot see them through the assistant either. And there is nothing new to learn: you just ask.

For IT

It runs on your environment

A small server cabinet with a single red cable plugged into the front.

Roles and permissions come from your own system. Whoever leaves the company drops out of the roles. Every request is in the log: who, what, when.

For the privacy question

You decide what may leave

A closed door with a letterbox, a red envelope halfway through the slot.

Only what someone asks for goes to the assistant, and what that can be is fixed per gate. Not your whole database, but that one answer.

Far-reaching actions end in a draft and a question, and a human clicks.

This only works if your system already does the work.

This fits you
  • Your team works every day in a system that already exists: a CRM, an order system, a schedule, an editorial environment.
  • There is a recurring task you could explain in steps to a new colleague.
  • You want people to use ChatGPT or Claude, but not to paste company data into it.
  • Someone on your side can say what is and is not allowed.
This does not fit
  • You are looking for a chatbot for your website, and that is a different kind of work.
  • The work still sits mainly in email, paper and loose files, not yet in one system.
  • You want the assistant to decide and publish on its own, and that is not how I build it.
  • You are looking for the cheapest standard package, while this is bespoke work on your system.

02 · Six steps, two prices

How does it work?

Design first, then build. The design is a separate decision point: only once you know which gates there will be and what they may do is building a sensible expense. Six steps, two prices.

A stopwatch with a red start button.
Intake · half a day

Which tasks cost the most time

Together we determine which tasks cost your team the most time, and which of those lend themselves to an assistant.

What you have afterwards · A list of tasks, ordered by hours per week, with whether a gate is possible for each.

A paper model of a wall with three small hatches, one of them red.
Design

The gates on paper

I work out the gates: what may happen, who may do it, what requires confirmation. You approve that before anything is built.

What you have afterwards · The design, even if you do not build further. You can stop here without losing anything.

A pipe fitting with a pressure gauge with a red needle.
Build and test

The connector on your system

On or in front of your system, with logging, roles and permissions. We test with your own people.

What you have afterwards · A working connector, tested with your own people on your own data.

A knife switch flipped on, with a red handle.
Go live

Rollout to the team

Short instruction, documentation, and the team starts working with it.

What you have afterwards · The documentation and the code are yours.

A thick logbook with a red ribbon bookmark.
Return visit · after four weeks

Walking through the log

Which gates are actually used, and which question keeps coming back without a gate for it.

What you have afterwards · A list of what could be added next, or the conclusion that it is done.

An oil can with a long spout and a red cap.
Safety net · ongoing

Optional maintenance

Updates when the standard changes, a fixed number of hours per month for new gates, and access to gates I already built for others.

What you have afterwards · Cancellable monthly. If you cancel, everything keeps running.

After every step you decide yourself whether to continue.

Want to see what that looks like for your team?

I run it on Careerguide’s real system. Five minutes, and no quote attached.

Request a demo

03 · Running in production at Careerguide

What can you check?

At Careerguide, publishing a vacancy went from nine steps to one question to an AI assistant, on the existing system and with the permissions that were already there. First the arithmetic, then what changed. Open a row for the detail.

30 min

per vacancy, by hand

Nine actions, spread across email, PDF and the form.

20 per week

arriving every week

Work you cannot get rid of. Every vacancy simply has to go in.

520 hours

of retyping per year

This is not difficult work, and that is exactly why it keeps getting put off.

A stopwatch with a red start button.
Before · by hand

Nine actions per vacancy.

Out of the email, through three screens, into the form. Every single time.

The nine steps
  1. 01 Pull the text out of the email or the PDF
  2. 02 Look up the organisation among the other clients
  3. 03 If it is not there yet: first create a new organisation
  4. 04 Find the right location
  5. 05 Choose field and segment
  6. 06 Tick experience, level, contract type and education
  7. 07 Rewrite the text so it can be found
  8. 08 Write a three-sentence summary for the vacancy card
  9. 09 Retype everything into the form and save

Half an hour in total. Twenty times a week.

A blue pipe connector with a red valve.
Now · in your AI assistant

“Publish this vacancy.”

One question with the PDF attached. The draft is in the system; you publish it yourself.

See the conversation
ChatGPT Claude Copilot
Careerguide · connected
PDF Vacancy — Financial Analyst.pdfPDF · 2 pages

Publish this vacancy.

Searched clients · Meesters & Partners found, Utrecht office, Maliebaan

Read it. Here is what I pulled out:

  • Role: Financial Analyst
  • Location: Utrecht
  • Organisation: Meesters & Partners, Maliebaan office
  • Contract: permanent, 32–40 hours

I rewrote the text for findability and wrote a three-sentence summary. Shall I show you those first?

Yes, and set the field to Finance.

Draft created in Careerguide · not yet published

Adjusted. The draft is in the system, not yet visible to visitors. You publish it yourself in Careerguide.

What you are looking at

The PDF simply travels along in the conversation, without a new screen and without retyping.

The assistant uses three gates on Careerguide’s system: search clients, rewrite text, create draft.

The assistant is not allowed to publish. That stays with a person, in the system itself.

Example conversation with example data. I show the real one live.

Example figures, not yet a measurement. During an intake we replace them with what actually happens.

04 · Price and next steps

What does it cost, and then what?

Two separate steps, each with its own price. After the intake you know what the design costs; after the design you know what the build costs, and only then do you decide.

Two separate steps, each with its own price.

What it costs

€2.500

the design · excl. VAT

After the intake you know what the design costs. After the design you know what the build costs, and only then do you decide.

  • The gates worked out: what may happen, who may do it, where a human approves.
  • A judgement on whether your system lends itself to this, and if not, why not.
  • Then the build, from €12.500: on your system, with logging, roles and permissions, tested with your own people. The price depends on the number of gates and is fixed before we start.
  • Optionally a safety net from €350 per month: updates, hours for new gates, cancellable monthly.
  • Code, design and documentation are yours, even if you do not build further. It runs on your environment, with your keys.
A pipe coupling with a red valve wheel on top.
Also have a look at
A scaffold of blue tubes with a single red plank on the top level.

Is it more than a connector? Then a Build begins: a fixed scope, a fixed price per phase, and after every phase you can stop.

See The Build
A drafting table with a rolled drawing and a red T-square.

Do you want someone to keep it sharp? Then I stay involved two to four days a month: assessing new gates, guarding the architecture, guiding your team.

See Fractional AI Architect

05 · What people want to know

Still have questions?

The questions that always come up, with short answers. If yours is not among them, a half-hour call is the quickest way.

Does this work with ChatGPT, or does it have to be Claude?

Both. The connector is independent of the assistant. You choose which one you use and you can switch later without rebuilding.

Do we have to move to a different system?

No. The connector sits on or in front of the system you already have. For the people working in it nothing changes, except that an assistant can now reach it through the door you open.

Can the assistant break something?

It can only do the actions in the design, with the permissions of the person asking. Far-reaching actions end in a draft and a question; a human clicks.

Our software is from a vendor. Is that possible?

That depends on whether there is an API, or one can be made. That is the first thing I look at during the intake. If there is none, you hear that straight away and it stops there.

Where does our data go?

The connector runs on your environment. Only what someone requests goes to the assistant, and you decide per gate what that can be. Every request is in the log.

How long before we get something out of it?

That depends on the number of gates. The intake is half a day, and after that you know what you are getting into. The design is a separate decision point, so you are not committed to anything until it is there.

Am I locked into anything?

No. One company, one connector, no subscription you cannot get out of. After every step you decide whether to continue.

Portrait of Daniel Plomp. A connector of two pipes with a red valve on top.
Who builds this

You work directly with the person who makes it.

I am Daniel Plomp. I have been building software since 1999, and in recent years mostly connectors between existing systems and AI assistants. The example above is one of them: it runs on Careerguide’s real system. There is no account manager and no intermediate layer: you explain your work once, to the person who then builds it.

One point of contact · no agencies in between

Five minutes, and you have seen it.

I show the conversation above live on Careerguide’s real system. After that you know which retyping on your side can go first. No quote attached.

Usually a reply within one working day