A tight two-minute Postman demo of driving an entire pull request from a single natural-language prompt. With the GitHub MCP server configured inside Postman's Agent Mode, one instruction — "create a branch, write tests for each API, commit, and request a review" — triggers a chain of GitHub tool calls that ends with a branch created, test scripts added, a PR opened, and a teammate assigned to review it.
The demo opens in Postman's Agent Mode and starts with configuration. Clicking the gear icon and choosing configure MCP servers reveals a GitHub MCP server already registered, along with the full list of tools it exposes. Each tool can be individually enabled or disabled, and as the presenter notes, there are a lot of options here. Connecting to the server does not require pasting a secret into the prompt: it relies on a personal access token stored securely in Postman's vault.
"when I connect to the GitHub MCP server, it's going to use my personal access token that's saved in vault."
Once connected, the same GitHub tools appear ready to use. You can click any of them and run it manually to see its results directly — a useful way to inspect behaviour tool by tool. But the point of the demo is to hand the orchestration to Agent Mode rather than firing each tool by hand. The setup for that is an AB testing collection that performs basic CRUD operations for an AB testing app, and which is already connected to the corresponding GitHub repository.
With the collection and repo linked, the presenter gives Agent Mode a single natural-language instruction: create a new branch in the AB testing repo called add tests, write tests for each API in the collection, add those tests to a commit, and then request a review from a teammate. That one sentence encodes the entire pull request lifecycle — branch, code, commit, PR, and review request.
"tell agent mode create a new branch in my AB testing repo called add tests. Write tests for each API in this collection. Add those tests to a commit and then request a review from one of my teammates."
Agent Mode then runs a burst of GitHub MCP tool calls. On screen you can watch the operations stream by: create branch, get file contents, push files, create pull request, and update pull request. Each is a discrete call into the GitHub MCP server, chained together automatically to fulfil the original request. There is no manual git checkout, no local commit, and no browser round-trip to open the PR — the agent handles the sequence.
When the run finishes, the workflow matches the prompt end to end: Agent Mode creates the branch, writes and adds the test scripts, and opens a pull request. Following the PR link shows the opened request, and the Files changed tab displays every test the agent authored. Crucially, the PR is already assigned to a teammate for review — the loop stops short of merging, keeping a human in control of the final decision.
"It created a new branch called add test, which is exactly what I wanted. It added the test scripts. It opens a PR."
"when I connect to the GitHub MCP server, it's going to use my personal access token that's saved in vault."
"It created a new branch called add test, which is exactly what I wanted. It added the test scripts. It opens a PR."
"And we can see it was assigned to my teammate for review."
| Approx. Time | Topic |
|---|---|
| 0:00 | Opening Agent Mode and configuring the GitHub MCP server |
| 0:17 | Enabling/disabling tools and connecting with a vault token |
| 0:33 | Running MCP tools manually vs. using Agent Mode |
| 0:43 | The AB testing collection and its connected GitHub repo |
| 0:57 | The prompt: branch, write tests, commit, request review |
| 1:09 | Agent Mode executes the GitHub MCP tool calls |
| 1:33 | Reviewing the result — new branch, test scripts, opened PR |
| 1:56 | PR assigned to a teammate for review |