Connecting an AI assistant to your project-management system used to mean giving it a better search box.

Ask what tasks are overdue. Summarize a project. Find a ticket. Explain what happened last week.

OpenProject 17.8 takes the next logical step: AI assistants connected through MCP can now write back.

The project’s MCP server can create and update work packages, add comments and manage relationships between tasks. That sounds like a relatively technical update, but it represents an important shift in what AI integrations can actually do for a small business.

Instead of asking an assistant what needs doing and then manually updating the project-management system yourself, the assistant can increasingly complete that administrative work for you.

Imagine running a small digital agency. After a client meeting, you give the transcript to your AI assistant and ask it to update the project. It identifies three new requests, creates the corresponding tasks, adds the client’s comments to an existing ticket and links one new task as dependent on another.

Or a developer finishes investigating a bug. The AI can update the work package with what was discovered, change relevant fields and add relationships to other affected tickets instead of someone manually copying everything across.

A project manager could ask: “Create tasks for everything agreed in today’s meeting, assign the correct project and add the relevant context to each ticket.”

That is where MCP becomes much more interesting for SMBs.

Reading information saves time finding things. Writing information starts eliminating work.

OpenProject first introduced its MCP server in version 17.2, allowing connected AI assistants to securely access project information. Version 17.8 expands that integration with create_work_package and update_work_package tools alongside support for comments, relationships and custom fields.

Importantly, those actions don’t bypass OpenProject’s existing controls. The MCP tools follow the same permissions and business rules as the normal interface and API. If a user doesn’t have permission to perform an action normally, connecting an AI assistant doesn’t magically give them additional authority.

That model is important as businesses start giving AI systems write access to operational software.

An assistant that can search Jira, a CRM or project-management system is useful. An assistant that can change those systems can save dramatically more time, but mistakes also become considerably more expensive.

For a small team, the potential productivity gain is obvious. Project administration often gets distributed across the people doing the actual work. Developers update tickets. Account managers copy client feedback. Project managers turn meetings into tasks. Owners chase everyone because half the information hasn’t been updated.

An agent capable of handling the mechanical parts could keep the system of record updated while humans focus on decisions and delivery.

The same pattern will extend far beyond OpenProject.

MCP started as a standardized way for AI systems to access external tools and information. As integrations mature, more servers will inevitably move from read-only operations toward creating records, updating systems and triggering workflows.

That is when MCP stops being primarily a convenient integration standard and starts becoming business automation infrastructure.

It also means businesses need to pay much more attention to permissions. Give an AI assistant the same principle you should give an employee: only the access required for the job.

OpenProject’s implementation is a useful example because the AI doesn’t get a special administrative back door. Existing permissions and business rules still apply.

For SMBs, that’s probably what useful agentic automation should look like.

Not an all-powerful AI running the company.

An assistant that can finally do the boring updates after you’ve told it what needs to happen.