As a software engineer, I hate context switching. Investigating a Jira ticket usually means bouncing between Jira for the context, Confluence for the high-level logic, the Codebase for the actual implementation, and finally Slack to ping the reporter about blockers/additional questions.
It takes time, breaks flow, and is incredibly tedious. So, I decided to automate the entire process using Claude Code and MCP.
I connected Claude to my Jira, Confluence, and Slack using MCP (Model Context Protocol).
Now, my day starts simply: I ask Claude for a list of my active Jira tickets and give it a command to investigate a specific one.
In about 2 minutes, Claude does the heavy lifting. It either prepares a complete implementation plan, or it compiles a list of blockers and clarifying questions that prevent me from moving forward.
I just start coding session If the plan is ready and there are no blockers. If there are blockers, Claude automatically drafts and sends a Slack message to the ticket reporter asking for clarification.
You can see the workflow diagram below:
- Everything starts with asking Claude: "Which tickets do I have in Jira?"
Claude uses the Jira Explorer Skill to hit the Atlassian MCP and pull my current tasks.
- Once the ticket is selected, I give the command: "Investigate this ticket and prepare an implementation plan."
This triggers the Knowledge Researcher Skill. The magic here is the strict order of execution. I explicitly force Claude to search Confluence first, and only then look at the Codebase. Confluence usually contains the high-level logic and API specs. If the agent jumps straight into the code, it scans blindly, wasting both time and tokens.
If all the information is there, Claude gives me a ready-to-use, step-by-step coding plan. But during the research, Claude might realize we are blocked (e.g., a missing payload structure or an undefined API endpoint). It will immediately stop and prepare a list of these blockers.
- Since I can't start coding, I tell the agent: "Find the reporter of this ticket and send these blockers to them on Slack."
Claude triggers the Slack Researcher Skill, finds the right person via Slack MCP, and drafts a highly professional message explaining exactly what is missing.
I’ll be honest, getting here wasn't without its challenges. I had to fine-tune the agent and figure out these strict rules through a lot of trial and error to prevent token drains and hallucinations. But right now, it works flawlessly.
Anyway, if you want to see how this actually works in reality, I recorded a step-by-step video on YouTube with real example:
If you want to skip the overview and example and dive straight into the exact prompts I wrote, and how I configured the custom skills and hard rules, you can jump directly to [05:22]