Claude
The fastest route: enter the connector, sign in, ask.
- On every plan, including Free (one connector)
- The person sets it up; on Team and Enterprise an owner does so centrally
- Claude asks before every tool call
Claude, ChatGPT or Microsoft Copilot: WORKSPACE.PM provides an MCP server through which any MCP-capable AI assistant connects to your projects. Each person signs in as themselves, every answer comes with exactly their permissions, the data stays in Germany. Which assistant you use is your decision, and you can change it at any time.
One endpoint, any assistant
Claude
connector, 2 minutes
ChatGPT
developer mode
Microsoft Copilot
Copilot Studio agent
WORKSPACE.PM MCP server
OAuth per person · rights 1:1 · role gate
Most PM tools sell their own AI copilot as an add-on module. WORKSPACE.PM has one too, but does not require it: the assistant your organisation has already approved gets access to the project data.
Whether your organisation has rolled out Claude Enterprise, ChatGPT Business or Microsoft 365 Copilot: the contract is signed, data protection is reviewed, staff are trained. WORKSPACE.PM builds on that instead of asking for a second AI decision.
No second AI provider, no second DPA, no second training programme.
Each person signs their assistant in as themselves. WORKSPACE.PM checks every call server-side with their roles and the tenant separation in the database. An additional role gate defines whether external AI clients may not work at all, read only or write.
The assistant never sees more than the person asking.
MCP is an open standard. If your organisation moves from one assistant to another, the connection to WORKSPACE.PM stays the same. Local models in your own data centre speak the standard as well.
The investment is tied to your project data, not to an AI vendor.
What differs is the route inside the respective assistant and who takes it. What stays the same is everything on the WORKSPACE.PM side.
The fastest route: enter the connector, sign in, ask.
Developer mode for custom connectors, enabled by the ChatGPT administrator.
For organisations in the Microsoft ecosystem: an agent in Copilot Studio.
Vendor details according to their public documentation as of September 2026. The WORKSPACE.PM rows apply to all three alike.
| Aspect | Claude | ChatGPT | Microsoft Copilot |
|---|---|---|---|
| Custom connector available on | Free (one connector), Pro, Max, Team, Enterprise | Plus, Pro, Business, Enterprise, Edu | Copilot Studio with Copilot licence or credits |
| Who sets it up | the person; on Team and Enterprise an owner | the person; on Business and Enterprise after admin approval | a maker builds the agent, IT approves |
| Effort to first answer | minutes | minutes | agent project, hours to days |
| Confirmation before tool calls | yes, per tool once or always | yes, write actions require confirmation by default | depends on the agent, not guaranteed |
| Central management of connectors | Team and Enterprise | Business and Enterprise | Microsoft 365 admin center, Power Platform DLP |
| Training on your chats | Team and Enterprise no; Free, Pro, Max on by default, can be turned off | Business and Enterprise no; individual plans via settings | no according to Microsoft |
| Permissions in WORKSPACE.PM | exactly the person's | exactly the person's | exactly the person's |
| Role gate for AI clients | no access, read only, write access | no access, read only, write access | no access, read only, write access |
| Audit trail | channel MCP with client name | channel MCP with client name | channel MCP with client name |
| Project data location | Germany | Germany | Germany |
Three steps, regardless of the assistant. Step one belongs to IT, the other two to the person.
In the WORKSPACE.PM Integration Hub, check MCP access and set the role gate for external AI clients. Recommendation for the start: read only, enable write access role by role.
Enter the MCP address in the assistant and sign in with your own WORKSPACE.PM account. The consent screen shows application, account, tenant and scope. Enforced SSO applies here as well.
From now on the assistant answers questions from the project data and, if the role allows it, carries out changes. Every change is in the audit trail with channel MCP and the client.
Regardless of which assistant is asking.
Every call runs under the person's identity, checked server-side with roles and row-level security. No service account, no shared token.
Per role: no access, read only or write access for external AI clients. The gate can only narrow permissions, never widen them.
Changes in the audit trail with channel MCP and client name, filterable by source. Read access is counted per person, not logged by content.
One tenant-wide switch disconnects all MCP access, running connections end immediately. Deactivated accounts lose access even with a valid token.
About the open connection in general. The details per assistant are on the respective pages.
MCP stands for Model Context Protocol, an open standard through which AI assistants access external systems. WORKSPACE.PM provides an MCP server offering over 200 project functions: projects, portfolios, risks, resources, Kanban, OKRs, requirements and more. With it, your organisation's assistant can answer questions from real project data and, where allowed, carry out changes.
Any client that speaks MCP over Streamable HTTP with OAuth. Documented are Claude (web, desktop, Cowork), ChatGPT in developer mode and Microsoft Copilot via an agent in Copilot Studio. Local models with an MCP-capable client can be connected as well.
No. The MCP server works without a configured AI provider. The embedded assistant is a separate, optional route with approval cards before changes and an optional model provider of your own. Both routes can be switched on and off independently.
No. Each person signs their assistant 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.
That is governed by the respective client: Claude asks before tool calls, ChatGPT requires confirmation for write actions by default, with Microsoft Copilot it depends on the agent that was built. Independently of that, WORKSPACE.PM decides through the role gate whether write access is possible at all, and logs every change with channel MCP.
Check the MCP switch in the Integration Hub and set the role gate for external AI clients, to read only for the start. The usage overview shows every connected person with requests, last use and client. Restricting to individual applications is not possible at present; whoever has access can use any MCP-capable client, always only with their own rights.
The project data stays in WORKSPACE.PM in Germany. The assistant retrieves answers, which become part of the chat history at the respective vendor, under its terms and your plan. WORKSPACE.PM sends nothing to an AI vendor on its own and stores neither questions nor answers.
Yes, provided the MCP endpoint of your installation is reachable from the internet over HTTPS, for example through a reverse proxy. Claude, ChatGPT and Copilot connect from the respective vendor's cloud. Purely internal installations without internet ingress can only connect local models.
Thirty minutes with someone who knows the MCP server from the code. We show the connection live and answer your IT's questions.
Product facts as of September 2026, taken from the WORKSPACE.PM source code and tool catalogue. Statements about Claude, ChatGPT and Microsoft Copilot are based on the public documentation of Anthropic, OpenAI and Microsoft as of September 2026. Claude is a trademark of Anthropic, PBC; ChatGPT a trademark of OpenAI; Microsoft and Copilot trademarks of the Microsoft group of companies. WORKSPACE.PM is not affiliated with these companies.