WORKSPACE.PM turns ChatGPT into a project assistant that knows your real data: status, risks, workload, portfolio. Connected through a custom connector in developer mode, with each person's own permissions, operated in Germany. For ChatGPT Plus, Pro, Business and Enterprise.
Campus North is amber. 2 of 5 milestones done, 3 open risks, of which 1 critical (facade delivery delay). Cost positions: 1.2 million planned, 0.9 million committed.
Open in WORKSPACE.PM
200+project functions ChatGPT operates for you, from the status question to a new risk
minutesfrom connector address to first answer, once developer mode is enabled
DEproject data stays in German data centres, ChatGPT pulls
What changes when ChatGPT knows your projects
In many organisations ChatGPT is the assistant staff already use. What it lacks is access to project reality: dates, risks, capacity, costs. The WORKSPACE.PM connector delivers them, current and within the permissions of the person asking.
Answers instead of status meetings
Executives and steering committees ask ChatGPT for the state of a project or the whole portfolio and get the answer in seconds, with a link to the source. The status round becomes a decision round.
Leaders no longer wait for the report. They ask.
Get project work done in the conversation
A risk surfaces while talking to ChatGPT? It creates it, measure included, after your confirmation. A capacity question before a commitment? It checks the assignments. Time entry, Kanban card, lessons learned: done without switching applications.
Fewer context switches, better maintained data, because maintenance happens on the side.
From a single seat to a managed organisation
On ChatGPT Plus and Pro the person sets up the connector. In ChatGPT Business and Enterprise the ChatGPT administrator enables developer mode, reviews the connector and publishes it for the organisation. On the WORKSPACE.PM side, IT decides through the role gate what external AI clients may do.
One route for the pilot, the same route for the rollout.
Six things you get done with ChatGPT from today
Real working situations, real answers from WORKSPACE.PM. Every answer links back to the object in the system.
Steering committee
Status without copying from three windows
Prompt
How is project Campus North doing: milestones, open risks, cost positions?
Result
ChatGPT pulls the figures from WORKSPACE.PM and answers with clickable links to the project, its risks and the latest report.
Create a risk in project Campus North: facade delivery delay, likely, high impact, measure: request second supplier.
Result
If the person's role has write access for AI clients, ChatGPT creates risk and measure after your confirmation. The audit trail records channel MCP and the client.
Custom connectors in ChatGPT run through developer mode. On Plus and Pro the person enables it; in Business and Enterprise the ChatGPT administrator does. After that the route is the same: enter the address, sign in, ask.
1
Enable developer mode
Turn on developer mode in the ChatGPT settings. In ChatGPT Business and Enterprise the admin must allow it in the ChatGPT organisation settings first; if the option is missing, that is the most common reason.
2
Create the connector
Under Apps and connectors create a new connector, give it a name, enter the MCP address of WORKSPACE.PM and choose OAuth for sign-in; your administrator finds the address in the Integration Hub.
3
Sign in, ask, confirm
ChatGPT opens the WORKSPACE.PM sign-in with the consent screen. From then on it answers questions from your project data; write actions require confirmation in ChatGPT by default. Write access exists only if your role allows it for AI clients.
What your IT needs for it
Check two settings in the WORKSPACE.PM Integration Hub: the MCP switch and the role gate for external AI clients, which we recommend setting to read only for the start
The MCP endpoint must be reachable from the internet, because ChatGPT connects from OpenAI's cloud; for on-premises, through a reverse proxy
For ChatGPT Business and Enterprise: the ChatGPT administrator allows developer mode and can publish the reviewed connector for the organisation
According to OpenAI documentation as of September 2026. WORKSPACE.PM needs no AI provider of its own and no second contract for this route.
Plan
Custom connector
Who sets it up
Training on your chats
ChatGPT Free
no
not possible
per OpenAI settings, on by default
ChatGPT Plus and Pro
yes, developer mode
the person
setting Improve the model for everyone, on by default, can be turned off
ChatGPT Business
yes, after approval
ChatGPT administrator allows, then the person
no, business terms
ChatGPT Enterprise and Edu
yes, centrally managed
admin or authorised developers, published by the admin
no, business terms
Recommendation for organisations: ChatGPT Business or Enterprise, because the admin reviews and publishes the connector there and chats are not used for training. Availability of developer mode and custom connectors can vary by region and ChatGPT organisation; OpenAI has rolled out connector features later in the EEA, Switzerland and the UK in the past. Check whether the option appears in your ChatGPT organisation.
For IT, CISO and data protection
ChatGPT works with your permissions. Not with more.
The connector is deliberately not a service account. Each person signs ChatGPT in as themselves, WORKSPACE.PM checks every call server-side, and your IT limits per role what external AI clients may do. What OpenAI does with the chat history is governed by your OpenAI plan.
Identity
ChatGPT never signs in as an application
Each person signs ChatGPT in as themselves, via OAuth 2.0 with PKCE. No service account, no shared token. The consent screen names the application, the account, the tenant and the scope.
If SSO is enforced, your identity provider's policies including MFA apply to this sign-in as well
An account deactivated in WORKSPACE.PM loses access immediately, even with a valid token
To make a block in your directory arrive automatically, enable SCIM 2.0 provisioning; without SCIM, block the account in WORKSPACE.PM as well
Sign in to WORKSPACE.PM
Application
ChatGPT
Account
m.lang@example.com
Tenant
Example AG
Scope
workspace:pm
The application acts with exactly your permissions.
CancelAllow
Permissions
Exactly the rights of the person, plus a gate for AI clients
ChatGPT acts with exactly the permissions of the signed-in person, checked server-side with roles and row-level security. On top of that you limit per role what external AI clients may do: nothing, read only or write.
The gate can only narrow permissions, never widen them
Our recommendation for the start: read only, then enable write access role by role
The gate applies to all external AI clients alike; restricting it to a single application is not possible at present
ChatGPT requires confirmation for write actions by default; whether a change is possible at all is decided by WORKSPACE.PM
External AI clients per role
No access
Read only
Write access
Administrator
PMO
Team
80 read tools0 write tools
The gate can only narrow permissions, never widen them.
Traceability
Every change carries channel MCP and the client name
Every change triggered through ChatGPT is written to the audit trail with channel MCP, the acting person and the name under which the client registered, linked to the changed object and filterable by source.
Administrators see which person connected when, how many requests ran and which client was used last
Read access is counted per person, not logged by content; WORKSPACE.PM stores neither your prompts nor the answers
The audit trail is append-only and records the channel, not the network address of the client
One switch disconnects the entire MCP access of the tenant, running connections end immediately
Connected people
PersonRequestsLast client
M. Lang142ChatGPT
S. Kern37ChatGPT
B. Bolzmann9Claude Desktop
Audit trail, filtered by source: MCP
Risk R-18 created
M. Lang
Channel: MCPClient (as registered): ChatGPT
Card K-204 moved
S. Kern
Channel: MCPClient (as registered): ChatGPT
Data flow and contracts
ChatGPT pulls. WORKSPACE.PM does not push.
ChatGPT connects from OpenAI's cloud to the MCP endpoint and retrieves data. WORKSPACE.PM sends nothing to OpenAI on its own and runs no AI model of its own for this route. There is no broker in between and no additional store.
Project data is processed in German data centres or in your own operation, governed by the data processing agreement under Art. 28 GDPR
Which data an AI client you connect retrieves is your decision as controller; OpenAI acts on the basis of your contract with OpenAI, WORKSPACE.PM is not involved in that processing
Answers become part of the ChatGPT chat history; whether OpenAI uses them for training depends on plan and settings, see the plan table
Depending on the question, answers also contain personal data of your employees, such as names, assignments and time entries; use the role gate to limit which roles get access at all
Your contract with OpenAI
ChatGPT
connector in your ChatGPT account
DPA with WORKSPACE.PM
WORKSPACE.PM
MCP endpoint, Germany
pulls with the rights of the person
answer becomes part of the ChatGPT chat history
no push, no broker, no extra store
Operated where your rules require it
The ChatGPT connection works the same in all three operating models. What changes is who runs the endpoint.
SaaS in Germany
Operated in German data centres, DPA under Art. 28 GDPR, no US sub-processor for project data.
Managed hosting
Your own instance, operated by us, in Germany or in the data centre you specify.
On-premises
Your own operation. The MCP endpoint of your installation must be reachable from the internet over HTTPS, for example through a reverse proxy.
Purely internal installations without internet ingress cannot connect ChatGPT. For public sector clients: on-premises and managed hosting in Germany, connection optional and switchable per role.
The four questions that come up in every security review, answered with what the product actually does.
Then ChatGPT sees everything?
No. Every answer is computed under the identity of the person asking. Confidential projects remain invisible, including for search and evaluation. What OpenAI stores from the answer is governed by your OpenAI plan.
What if the AI changes something wrong?
Write tools can be blocked completely per role or limited to reading. ChatGPT requires confirmation for write actions by default; in the system the person's normal rights apply in addition. Every change is marked as an MCP change in the audit trail.
What does the works council need to know?
WORKSPACE.PM stores no questions you ask ChatGPT and no answers. Stored per person are the number of requests, the time of last use and the client, plus every change with channel MCP. The chat history lives with OpenAI, under the rules of your plan.
Are we locking ourselves into OpenAI?
MCP is an open standard. The same endpoint works with Claude, Microsoft Copilot or local models. If you change the assistant, the connection stays. And you can switch the integration off tenant-wide at any time.
Frequently asked questions
Short answers for project leads, IT and data protection.
Yes, through a custom connector in ChatGPT's developer mode that uses the MCP server of WORKSPACE.PM. The person enters the MCP address and signs in with their WORKSPACE.PM account. According to OpenAI this is available on ChatGPT Plus, Pro, Business, Enterprise and Edu on the web; not on the Free plan.
On the WORKSPACE.PM side, MCP access is part of every tenant, without an add-on licence and without an AI provider of your own. On the OpenAI side a paid plan from Plus upwards is required; the Free plan has no custom connectors. The usage limits of your ChatGPT plan continue to apply.
Read and, if the role allows it, write: projects, portfolios and scenarios, risks with measures, Kanban cards, time entries, resource assignments, OKRs, requirements, lessons learned and more, over 200 tools in total. Every answer links to the object in WORKSPACE.PM. Not reachable via MCP are meetings, documents, stakeholders, change requests, dashboards and user administration.
No. Each person signs ChatGPT in as themselves, and every call is checked server-side with exactly their rights, including tenant separation in the database. An additional role gate for external AI clients can only narrow these rights further, never widen them.
Yes. According to OpenAI, write actions in developer mode require confirmation by default; the decision can be remembered per conversation, and a new conversation asks again. Independently of that, WORKSPACE.PM decides through the role gate whether write access is possible at all, and logs every change with channel MCP.
In ChatGPT Business and Enterprise the ChatGPT administrator must allow developer mode in the ChatGPT organisation settings first; that is the most common reason for the missing option. Availability of developer mode and custom connectors can also vary by region; OpenAI has rolled out connector features later in the EEA, Switzerland and the UK in the past. Check OpenAI's current information for your ChatGPT organisation.
In WORKSPACE.PM through the tenant-wide MCP switch and the role gate for external AI clients, plus the usage overview with person, requests, last use and client. In ChatGPT Business and Enterprise additionally through the ChatGPT administrator, who allows or blocks developer mode and publishes reviewed connectors for the organisation.
WORKSPACE.PM sends nothing to OpenAI; ChatGPT retrieves answers, which then become part of your chat history at OpenAI. For Business, Enterprise and Edu, OpenAI's business terms apply without training on your data. On Plus and Pro the setting Improve the model for everyone controls training; it is on by default and can be turned off. Check OpenAI's current terms.
WORKSPACE.PM processes project data exclusively in Germany, as SaaS, managed hosting or on-premises. ChatGPT connects from OpenAI's cloud to the endpoint; there is no broker and no additional store. What OpenAI processes from the answer is governed by your contract with OpenAI, not by WORKSPACE.PM.
Yes. The same MCP endpoint works with every client that speaks Streamable HTTP and OAuth. Claude and Microsoft Copilot have their own pages with the respective routes. The overview of all assistants is on the page Connect AI assistants.
Thirty minutes with someone who knows the MCP server from the code: identity chain, role gate, audit trail and what your data protection officer will ask.
Product facts as of September 2026, taken from the WORKSPACE.PM source code and tool catalogue. Statements about ChatGPT are based on OpenAI's public documentation as of September 2026. ChatGPT is a trademark of OpenAI. WORKSPACE.PM is not affiliated with OpenAI.