Connect your AI assistant to TaskRay once, and you can ask about your projects in plain language instead of clicking through to find the answer. This guide is a set of ready-to-try prompts for real situations. This first article covers what TaskRay MCP is, what it can reach, and how it protects your data, so the rest of the guide makes sense before you run a single prompt.
In this article:
- What TaskRay MCP is, and where it works
- What it can and cannot access
- The two guardrails that keep your data safe
- Habits that get you more out of it
- Where to go next
What TaskRay MCP is
TaskRay MCP is a connection between your AI assistant and TaskRay. Once it is set up, you can ask your assistant to look things up, summarize a project, create tasks, log time, and more, all in plain language. The same work you have always done in TaskRay becomes something you can just ask for.
It works in any connected AI surface, including Claude, ChatGPT, Microsoft Copilot, and Agentforce. The prompts throughout this guide work the same way across all of them.
What it can and cannot access
TaskRay MCP is an interface to the TaskRay capabilities you have already licensed. It works with what you own, and it does not expand it. It does not grant new licenses, unlock additional editions or add-ons, or work around your Salesforce permissions. What your assistant can see and do depends on the features enabled for your organization and the permissions of the connected user.
Because access follows your own license and permissions, the more TaskRay capability your organization owns, the more your assistant can do. Each prompt in this guide lists what it needs, so you know before you try it.
The two guardrails that keep your data safe
Two things are worth knowing before you start.
The assistant acts as you. It can see and change only what your own Salesforce permissions allow. It holds no access of its own, so it can never reach data or actions you could not reach yourself in TaskRay.
It asks before anything sweeping. Deletes and bulk changes across many records do not run on their own. They are queued for you to review and approve first, and nothing changes until you approve it. You get the speed of one instruction with a human sign-off on the actions that carry weight.
Single, everyday edits (renaming a task, moving one card, logging an hour) happen right away. The review step is reserved for the actions with a wider reach, like deleting records or updating many at once.
Habits that get you more out of it
A few small habits separate teams that dabble from teams that compound.
- Name things precisely. "The Acme onboarding project" beats "that project we started." Your assistant is good at matching, but an exact name removes a round trip.
- Ask to see the list first. "Show me the list before you file them" is a good habit for anything that touches many records, and the approval step backs you up on the big ones regardless.
- Ask for the format you want. "Three bullets for slides," "a table," or "one paragraph for Slack." You will get exactly that.
- Chain your prompts. One prompt can find a problem, the next can explain it, and a third can fix it, all in the same conversation.
- Put the repeats on a schedule. Anything you run every Monday can run before you reach your desk. See the automations article for how.
Where to go next
- Choosing Your AI Model and Effort Level for TaskRay MCP, for picking the right model for the job.
- TaskRay MCP Requirements and Licensing, for what each prompt needs.
- First Prompts to Try After Connecting TaskRay MCP, to prove the connection and see what is possible.