Skip to content
GitHub Copilot Train-the-Trainer
Trainer Guide

Session M7-03 trainer guide: Plan work with GitHub Issues and MCP

Duration: 1 hour

Delivery objective

Show learners how to plan a GitHub MCP write, approve it, and check what GitHub stored. This continues the Delivery Decision Studio path. Once created, the live issues become the source of record.

One-hour plan

Time Segment
0:00-0:08 Connect the decision brief to saved GitHub issues
0:08-0:18 Explain MCP from the learner's point of view
0:18-0:30 Inspect tools, permissions, and repository context
0:30-0:43 Decompose the initiative into an issue plan
0:43-0:53 Preview, approve, and check a write
0:53-1:00 Explain the lab guardrails

Required preflight

  • Check that every learner can open the GitHub Copilot app.
  • Check that the approved GitHub MCP server is available.
  • In the app, open Customize → MCP → Installed and inspect the connection.
  • Test one GitHub read before demonstrating a write.
  • Check that learners can read the training repository and create issues in it.
  • Use a repository created for training. Do not use a production backlog.
  • Record the repository owner and name before the demonstration.

Important

Stop if the learner cannot identify the target repository, inspect the tool call, or check the resulting issue.

Explain MCP without implementation detail

Use this definition:

An MCP server gives Copilot a set of named tools for reading or changing another system.

Explain these rules:

  • the tool runs with the permissions available to the connection;
  • a read operation can still expose restricted information;
  • a write operation must name its target and proposed values;
  • Copilot can propose an action, but the learner approves it;
  • check the saved result with a new read; Copilot's success message is not enough.

Do not explain server protocols, transports, or SDKs.

Demo flow

1. Read the approved source

Open lab/starter/approved-decision-brief.md. Ask Copilot to identify the outcome, constraints, non-goals, owners, and results needed for acceptance.

2. Inspect GitHub

Ask Copilot to:

  • identify the target repository;
  • list existing labels relevant to the plan;
  • search for duplicate or overlapping open issues;
  • report any permission or context uncertainty.

Use lab/starter/existing-issues.json when the training repository does not already contain the seeded records. Have learners decide what to do with each candidate:

  • reuse #241 because it partially overlaps the requester view;
  • leave #256 unchanged because notifications are out of scope;
  • reject #263 as implementation work because it conflicts with the visibility boundary;
  • link #278 only as historical context.

3. Propose the issue set

Use one parent issue and four child roles: public status mapping, requester status view, empty and stale states, and requester-access verification. These roles feed Sessions M7-04 and M7-05. A fifth child is optional and stays outside those four workstreams. Each issue should have:

  • a clear outcome;
  • defined scope;
  • acceptance criteria;
  • a non-goal;
  • dependencies;
  • owner or ownership gap;
  • link to the decision brief.

Reject technical subtasks that have no independent outcome or review point. The Markdown issue plan is a temporary preview. Do not commit or hand it to the next session.

4. Preview the write

Before Copilot calls a write tool, ask it to show:

  • repository owner and name;
  • issue count;
  • each title and body;
  • labels and milestone;
  • parent-child relationship method;
  • fields that remain unset.

The learner records Approve, Revise, or Reject before the write.

5. Check and correct

After creation, retrieve the issues in a fresh read and compare them with the approved plan. Demonstrate a correction by adding the verification owner and source-decision path to one child issue. Preview the update, get approval, write it, and read it back.

Check that the live parent issue links the complete child set. Session M7-04 reads those issues directly.

Common failure modes

Failure Trainer response
Copilot targets the wrong repository Reject the operation and restate the repository
Duplicate issue exists Link or update the existing issue instead of creating another
Partial overlap exists Reuse the work already covered and create only the missing outcomes
Child issue is too broad Split it before the write
Labels do not exist Use approved existing labels or leave the field unset
Tool output claims success without a returned issue Run a fresh repository read
Learner cannot review tool arguments Stop the lab

Product notes verified October 5, 2026

Check current GitHub MCP availability, authentication, tool names, issue hierarchy support, and write confirmations before delivery.

Official references: