teamspace is a cloud ERP for professional services firms with its own MCP server and built-in AI features. Through the MCP connection, Claude, ChatGPT or Cursor work directly with your projects, times, tickets, contacts and invoices; inside teamspace itself, the AI condenses long tickets and reads reports. Both run on your own AI and only with the rights of the signed-in user, in every edition.
What AI in your ERP brings
An AI is only as useful as what it knows about your work. Inside the ERP it knows your projects, tickets, contacts and appointments, without anyone exporting lists or pasting text into some outside chat window. Day to day, that takes reading, searching and piecing things together off your plate.
Forty posts, three people, two weeks of back and forth: the AI sums up the history in a few sentences, with decisions, open questions and the current state. Useful for handovers, after a holiday, before you call the client and during an escalation, when someone has to get up to speed without the long backstory.
Instead of checking line by line, you let the AI read the report on screen. It points out what stands out, such as a project well over plan or a client whose revenue is falling away. The figures in the report remain what counts.
Through the MCP server you simply ask in Claude or ChatGPT: "Which tickets for Müller GmbH are still open?", "What is in my agenda this week?" or "Summarise what I worked on last month." The answer comes from your teamspace, provided a suitable MCP connection has been enabled for you, and it only shows what you are allowed to see there yourself.
Learn moreAfter the weekly meeting you tell your assistant: "Take the last minutes from the marketing meeting, drop items 2, 5 and 6 and add the autumn trade fair." Through a separate MCP connection for minutes it creates the new Markdown file in the teamspace file store. This requires that the connection is enabled for you and that you have write rights for that folder in teamspace itself.
Learn moreAnyone using AI today often pastes client data into whatever chat is to hand. With AI in your ERP, your own approved AI works on current data, and it only sees what the signed-in person is allowed to see.
Learn moreYour AI, your data
Many software vendors handle AI the same way: they sign a contract with a large model provider, build that model in and switch it on for every customer. As a customer, you are left not quite knowing where your data is processed, or on what terms.
teamspace takes the other route. We do not run a language model of our own. Each tenant enters its own AI access: endpoint, model and key. That way your organisation answers the question of where the data goes, not us. Processing happens with the provider you have a contract with yourself, or on your own server. No additional data processing arrangement arises through teamspace.
Your tenant data is not used to train AI models. That applies to today's AI features and to every one still to come.
AI-ready ERP
"AI-ready" now appears in almost every brochure. Six questions separate the label from what holds up in day-to-day work. Next to each one is how teamspace answers it.
An AI assistant needs controlled access to the ERP, otherwise it is left guessing. teamspace comes with its own MCP server for that, in every edition.
Learn moreA built-in model ties you to the vendor's choice. In teamspace you enter your own AI and switch it by changing one field.
It must not. In teamspace the signed-in user's rights apply to every AI call, and that cannot be switched off.
Not by teamspace, neither today nor in future features. Whether your own AI provider stores data is governed by your contract with that provider.
Learn moreInvoices and key figures have to come out the same every time. In teamspace rules do the calculating; the AI may check and flag anomalies.
An open API turns the ERP into a foundation for your own tools. With a coding agent you build small apps on your data; the REST API is available from enterprise.
Learn moreThree routes
The built-in AI features work inside teamspace itself. From outside there are three routes. All three use the same data record; they differ in who is asking and how often.
What always applies
Whatever feature is added to teamspace, these four points stay the same.
There is no teamspace AI. Without a connected endpoint, no AI feature appears at all. That is the default.
No background process transfers data and your tenant is never indexed in advance. Every call is something a person does, and only what the feature needs leaves the system.
No AI feature closes a case, writes an invoice or settles anything you have to answer for.
An AI feature never shows you anything you could not see anyway. The same holds for any AI assistant you connect through the MCP server.
Two routes
People call both "AI in teamspace". They mean different things, and the difference decides who sets up what.
AI features in teamspace
Your assistant via MCP
AI features in your ERP
teamspace uses AI where it makes a visible difference and where a mistake shows up rather than hiding. Both features only appear once your tenant has an AI of its own connected and your administrator has enabled them. They are included in every edition.
The action that generates the ticket subject turns a ticket with forty entries into a few sentences, organised in four parts: overall picture (what is this about?), decisions (what has already been agreed?), open questions (what still needs settling?) and current status. The summary goes into the ticket's subject field and replaces the text that was there; the individual entries stay unchanged.
In a report's three-dot menu, the AI analysis sits right below Export. It adapts to the kind of report: on a timesheet it thinks about plan versus actual; on an invoicing analysis, about revenue spread and clients dropping away. The result is organised into overall picture, anomalies, trend and questions worth a closer look. The AI sees exactly the data you have on screen and writes nothing back. Its pointers are an assessment, not a verdict; the report remains authoritative.
A ticket dispatcher that routes incoming tickets to the right channel or person, a ticket prioritiser, and a CRM data fetcher that fills in details on organisations. The descriptions may still change before release.
AI and rules
You could build business software today in which an AI writes the invoices. It would be right most of the time. For an invoice, though, "most of the time" is not enough: across a thousand invoices a month, one would eventually be wrong, it would look just like the others, and nobody could explain why that one.
That is why invoices, billing and figures in teamspace are produced by rules. The rules are stored, traceable and run the same way every time. If a rule is wrong, you find it and fix it, and from then on it is right for good. The AI may take a look and ask: "This project has been billed for three months without the framework agreement discount, although the client has one. Sure?" Whether it has a point is for a person to decide.
Where AI beats any rule, teamspace is glad to use it: condensing long histories, spotting anomalies in a report and explaining things in full sentences. These are tasks where you can judge the result at a glance.
Setting up
An administrator decides once. The connection applies to the whole tenant, not per user or per module, and from then on users simply see that the features are there.
An AI provider you have a contract with, or a model you run yourself.
Does the provider train on your data, how long does it store it, where are the servers? Those are questions for your contract.
Under Configuration, Interfaces, Actions, AI integration: endpoint, model, key (optional for a model you host in-house) and a maximum input length. It limits how much a single call transfers, and with it cost and data volume. Around 50,000 characters is typical; check that this suits your model.
Your administrator enables what your organisation wants to use. The Enabled switch turns every AI feature off at once.
Short call
A provider in the EU, your own model in your data centre, or just the MCP server for now: in a first call we go through what your data protection rules demand and how teamspace fits.
Making the call
The shortest way to put it: the API is for software that always does the same thing. MCP is for questions you do not know in advance.
REST API
MCP server
Both together
In practice the two routes do not exclude each other. The REST API keeps the nightly handover to accounting running, reliably and the same way every time. The MCP server answers the afternoon question of why the figures from that handover look odd.
A proven approach is to start with MCP. Once it turns out that a particular question is asked every week and the answer always comes about the same way, that is the moment to build it as a report in teamspace or as an integration via the API. For customers on the office edition, the MCP server is often the first way to connect teamspace to other tools at all, because the REST API is only included from enterprise.
For figures that must be exact and repeatable, the report remains the reliable source: an assistant words its answer afresh every time. How teamspace handles your data overall is set out on the security and data protection page.
First call
Let us show you AI on an example of your own.
Bring a long ticket or a report you would like to get to grips with faster. In the call you will see what the AI makes of it, and which data leaves your organisation in the process.
In a first call we work out which route suits you: built-in AI features, the MCP server for your assistants, or an application of your own via the API.
