Building a developer knowledge base
for GitHub Copilot

Pamela Fox

Principal Cloud Advocate

Microsoft / GitHub

Said Nikjou

AI Forward Deployed Engineer

Microsoft

Pamela Fox

Principal Cloud Advocate

Principal Cloud Advocate at Microsoft, focused on Python, AI, and developer tools. Previously, she lectured at UC Berkeley, and worked for Coursera, Khan Academy, and Google.

Said Nikjou

AI Forward Deployed Engineer

AI Forward Deployed Engineer at Microsoft, focused on AI agents, GH Copilot, Azure AI, and enterprise automation. He has spent over 15 years helping organizations turn technology innovation into business outcomes.

About you

Turn to your neighbor and introduce yourself:

Name
Location
What coding agents do you usually use?
What's your favorite LLM to use with the agent? What's your LEAST favorite?

Imagine joining a new team.

How do you get your questions answered?

How do you make sure you're building the right thing the right way?

The answers to your engineering questions
may require many sources

Code

What behavior exists today?

Evidence: implementation, tests, configuration

GitHub activity

What changed, why, and what remains open?

Evidence: PRs, issues, discussions

Documents

What does the business require or prohibit?

Evidence: requirements, ADRs, policies

Today: Give GitHub Copilot access to each of these sources progressively and see answer quality improve as you equip it with more sources.

Today's agenda

Time	Format	Activity
1:00–1:05	Welcome	Settle in and introductions
1:05–1:12	Presentation	Meet the project and your coding agent
1:12–1:27	Exercise	Explore the local code
1:27–1:32	Presentation	Connect Copilot to GitHub with MCP
1:32–1:45	Exercise	Add GitHub context
1:45–1:55	Presentation	Explore and connect a document knowledge base
1:55–2:10	Exercise	Add document context
2:10–2:15	Presentation	Use the combined knowledge base
2:15–2:25	Exercise	Use a multi-source knowledge base
2:25–2:30	Presentation	Take the pattern back to your team

The company: Cocoarynth

A fictional bean-to-bar chocolate company built around transparent sourcing.

Makes small-batch, single-origin chocolate
Works directly with cacao producers
Tracks every origin lot and production batch
Connects finished products to wholesale deliveries

The project: Cocoarynth Trace

A web app that follows every batch from origin to shipment

Lists production batches and their status
Shows cacao origin and production history
Associates batches with wholesale shipments
Generates an Origin Passport

Cocoarynth Trace chocolate batch list showing products, origins, production dates, statuses, and shipment state

The feature: Origin Passport

A machine-readable document that lets a wholesale retailer connect a chocolate bar batch to its origin and production history.

Cacao origin

→

Production batch

→

Origin Passport

→

Wholesale delivery

Includes approved facts

Product, batch, origin, producer, harvest, certification, production, and shipment details.

Leaves out confidential data

Supplier pricing, personal contacts, internal quality notes, and employee comments.

GitHub Copilot: Choose your client

GitHub Copilot App

Work from a project session connected to the repository.

GitHub Copilot CLI

Work from a terminal in the local repository.

VS Code

Work with the GitHub Copilot chat agent while you have repository open.

Use the same client for every exercise so you can compare across stages.
If you have time after using chosen client, you can experiment with the other two.

Exercise 1: Explore the local code

Your question

Using the implementation files and tests, explain what this project does, identify its main components, and describe how it exports an Origin Passport. Cite the files that support your answer.

The evidence

Local codeImplementation files
what exists

Local codeTests
expected behavior

🔗https://tinyurl.com/ghu-kb-workshop : exercise1.md

Connect GitHub Copilot to
the GitHub MCP server

Bring PRs, issues, and discussions into the context window.

MCP: Model Context Protocol

MCP is the open standard that tells agent runtimes how to discover and connect to the tools they can call.

🔗MCP specification: modelcontextprotocol.io

MCP architecture

Configure GitHub MCP in Copilot App

Open Customize → Add → MCP Server.

Name it github, select HTTP, and enter this URL:
https://api.githubcopilot.com/mcp/readonly

To enable more tools, add this header:
X-MCP-Toolsets
repos,issues,pull_requests,discussions

Copilot App Add MCP Server dialog with server name github, HTTP server type, URL https://api.githubcopilot.com/mcp/readonly, and header X-MCP-Toolsets set to repos,issues,pull_requests,discussions

Configure GitHub MCP in Copilot CLI

GitHub MCP is built in with read-only tools enabled by default.

1. Start a new session

Add the Discussions toolset:

copilot --add-github-mcp-toolset discussions

2. Verify the server

From the active Copilot CLI session:

/mcp show github-mcp-server

Confirm tools are available for repositories, issues, pull requests, and discussions.

Configure GitHub MCP in VS Code

1. Add `.vscode/mcp.json`

{
  "servers": {
    "github": {
      "type": "http",
      "url": "https://api.githubcopilot.com/mcp/readonly",
      "headers": {
        "X-MCP-Toolsets": "repos,issues,pull_requests,discussions"
      }
    }
  }
}

2. Start and verify

Select Start above the server definition.

Authorize with the event-provided GitHub account.

Confirm the github server is running.

Open Copilot Chat tools and find the GitHub tools.

Exercise 2: Add GitHub context

Your question

Use GitHub MCP to investigate this repository. Why was the Origin Passport export built this way, and what related work remains open? Cite the relevant pull request, issue, and discussion.

The evidence

GitHub MCPMerged PR
original choice

GitHub MCPOpen issue
remaining work

GitHub MCPDiscussion
field policy

🔗https://tinyurl.com/ghu-kb-workshop : exercise2.md

Connect GitHub Copilot to
a document knowledge base

The pilot: Mapayça Markets

Mapayça Markets is a grocery store that wants to sell Cocoarynth chocolate bars.

The pilot request

Cocoarynth will supply its Ecuador Esmeraldas 72% chocolate bar.

Each shipment needs an Origin Passport that Mapayça's inventory system can import without manual conversion.

The question: Can Cocoarynth Trace support the pilot today?

The missing context lives in documents

First page of the Mapayça Markets Traceability Pilot Requirements document

Mapayça pilot requirements

Defines the required CSV format, schema, dates, and acceptance criteria.

First page of the ADR-007 Immutable Origin Passport Snapshots document

Architecture decision

Explains why an Origin Passport is an immutable snapshot.

Data-sharing policy

Separates shareable traceability facts from confidential data.

Support runbook

Guides validation, regeneration, and import troubleshooting.

Foundry IQ ingests documents into a search index

A cloud pipeline turns source files into searchable, citable chunks.

Blob Storage

Stores a retrievable copy of each source document.

Search indexer

Detects additions and updates, then runs the ingestion pipeline.

Content Understanding

Extracts text, tables, images, structure, and metadata.

Chunk + vectorize

Creates semantic chunks and computes embeddings.

Search index

Stores text, vectors, metadata, and source references together.

For this workshop, the ingestion pipeline has already run.

Content Understanding preserves more than plain text

Rich extraction

Reads text and OCR, preserves tables as structured HTML, and describes images and diagrams.

Semantic chunking

Splits on meaningful document boundaries instead of arbitrary character counts.

Grounding metadata

Keeps document identity, page location, and source references with every chunk.

Source document

→

Structured extraction

→

Semantic chunks

Hybrid search improves relevance

Foundry IQ combines exact-term matching with semantic similarity, then reranks the fused results.

QueryWhat are the CSV export requirements?

→

Keyword searchFinds exact terms such as CSV, field names, and identifiers.

Vector searchFinds passages with similar meaning even when wording differs.

→

Reciprocal Rank FusionCombines both ranked result lists.

→

Semantic rerankingPromotes the chunks most relevant to the full question.

Search the document KB over MCP

Each Foundry IQ knowledge base exposes one MCP endpoint with one retrieval tool.

1. Endpoint


<search>.search.windows.net
  /knowledgebases
  /cocoarynth-kb-docs
  /mcp

One URL per knowledge base.

→

2. Tool call


{
  "name": "knowledge_base_retrieve",
  "arguments": {
    "query_variants": [
      "CSV export requirements",
      "Mapayça import format"
    ]
  }
}

Copilot writes the queries.

→

3. Result


[
  { "ref_id": "0",
    "content": "…CSV…" },
  { "ref_id": "1",
    "content": "…pricing…" }
]

Includes matching chunks, plus one reference block per ref_id.

The workshop portal

A window into the Foundry IQ indexes and knowledge bases behind this workshop.

Browse documents

See the source files in each corpus.

View document chunks

See the extracted passages and page metadata.

Search the index

Run hybrid search and compare ranked passages.

Cocoarynth Trace workshop portal showing project requirements, engineering practices, and combined knowledge base navigation

Exercise 3: Add document context

Your question

Can Cocoarynth support Mapayça Markets' traceability import today? Identify the blocker and related open work, then cite the pilot requirements and current implementation evidence.

The evidence

Foundry IQPilot requirements
retailer needs

Local codeImplementation
current behavior

GitHub MCPOpen issue
related work

🔗https://tinyurl.com/ghu-kb-workshop : exercise3.md

Connect GitHub Copilot to
a multi-source knowledge base

Engineering practices live in a second corpus

First page of the API Design and Export Contracts Guide

API and export contracts

CSV serialization, validation, and errors.

Engineering culture

Review, testing, rollout, and monitoring.

UI and accessibility

Keyboard, status, and error states.

React and TypeScript

Frontend implementation and tests.

Foundry IQ knowledge base

An agentic retrieval engine that can query across multiple sources.

Retrieval planning in action

Input

What product requirements and company engineering standards should guide CSV Origin Passport support?

→

Planned queries

Project requirements

CSV Origin Passport product requirements support
Origin Passport CSV requirements
CSV support passport requirements

Engineering practices

CSV support standards accessibility API frontend
React frontend engineering standards forms file upload CSV
engineering standards data import CSV accessibility

→

Output

8 references, merged and reranked

3.94Pilot requirements · “Traceability Pilot Requirements…”
3.38Pilot requirements · “Example CSV: origin_passport_id…”
2.48Architecture decision · “An Origin Passport is an externally shareable statement…”
2.46Data-sharing policy · “Origin Passport baseline…”
2.42API design guide · “Generate CSV with a standards-based library…”

One combined KB, or two MCP servers?

	Two MCP servers	One combined KB
Servers	`cocoarynth-kb-docs` `cocoarynth-kb-engineering-practices`	`cocoarynth-kb-all` over both indexes
Source choice	Copilot decides which server to call	Foundry IQ plans which index to search
Parallelism	Depends on the agent and model	Queries always run in parallel
Merging	Copilot merges the evidence itself	Foundry IQ uses specialized model for result merging and reranking

One combined KB gives every agent parallel retrieval and Foundry IQ reranking.

Exercise 4: Use multi-source knowledge base

Your question

Create a grounded PRD using evidence from direct GitHub MCP and cocoarynth-kb-all. Distinguish required behavior, current implementation, and recommendations, then cite the sources.

The evidence

Foundry IQPilot requirements
required behavior

Foundry IQEngineering practices
team standards

Local codeImplementation
current behavior

GitHub MCPIssues + PRs
planned work

🔗https://tinyurl.com/ghu-kb-workshop : exercise4.md

Next steps

Continue after the workshop

This week

Use the workshop environment

Workshop GitHub account: take your login credentials home
Workshop portal URL
Secrets doc with endpoints and key

After this week

Deploy your own

Provision the Azure infrastructure in your own subscription
Ingest documents and create the knowledge bases
Connect Copilot with your own GitHub account

Coding agents need context beyond code

Local code

What exists

Available in the local file system

GitHub MCP

What changed, and why

Available over an authenticated MCP server

Foundry IQ

What the organization needs

Searchable only after indexing into a knowledge base

Connect existing MCP servers for live systems.
Build indexed knowledge bases for unstructured documents.

More MCP servers to connect

Azure DevOps Microsoft

Work items, repos, pipelines, and wikis.

🔗github.com/microsoft/azure-devops-mcp

Work IQ Microsoft

Microsoft 365 email, meetings, and files.

🔗github.com/microsoft/work-iq

Linear Third party

Issues, projects, cycles, and comments.

🔗linear.app/docs/mcp

Slack Third party

Messages, channels, threads, canvases.

🔗docs.slack.dev/ai/slack-mcp-server

🔗Find more in the GitHub MCP Registry: github.com/mcp

Learning resources

GitHub Copilot

Foundry IQ

🔗Build smarter agents faster with unified knowledge & serverless retrievalaka.ms/FoundryIQ
🔗Improve recall by up to 54% with knowledge basesaka.ms/foundryiq-evals
🔗What is Foundry IQ?learn.microsoft.com

Building a developer knowledge base for GitHub Copilot

About you

The answers to your engineering questions may require many sources

Code

GitHub activity

Documents

Today's agenda

The company: Cocoarynth

The project: Cocoarynth Trace

The feature: Origin Passport

Includes approved facts

Leaves out confidential data

GitHub Copilot: Choose your client

GitHub Copilot App

GitHub Copilot CLI

VS Code

Exercise 1: Explore the local code

Connect GitHub Copilot to the GitHub MCP server

MCP: Model Context Protocol

MCP architecture

Configure GitHub MCP in Copilot App

Configure GitHub MCP in Copilot CLI

1. Start a new session

2. Verify the server

Configure GitHub MCP in VS Code

1. Add .vscode/mcp.json

2. Start and verify

Exercise 2: Add GitHub context

Connect GitHub Copilot to a document knowledge base

The pilot: Mapayça Markets

The pilot request

The missing context lives in documents

Mapayça pilot requirements

Architecture decision

Data-sharing policy

Support runbook

Foundry IQ ingests documents into a search index

Blob Storage

Search indexer

Content Understanding

Chunk + vectorize

Search index

Content Understanding preserves more than plain text

Rich extraction

Semantic chunking

Grounding metadata

Hybrid search improves relevance

Search the document KB over MCP

The workshop portal

Browse documents

View document chunks

Search the index

Exercise 3: Add document context

Connect GitHub Copilot to a multi-source knowledge base

Engineering practices live in a second corpus

API and export contracts

Engineering culture

UI and accessibility

React and TypeScript

Foundry IQ knowledge base

Retrieval planning in action

One combined KB, or two MCP servers?

Exercise 4: Use multi-source knowledge base

Next steps

Continue after the workshop

Repository

Use the workshop environment

Deploy your own

Coding agents need context beyond code

What exists

What changed, and why

What the organization needs

More MCP servers to connect

Azure DevOps Microsoft

Work IQ Microsoft

Linear Third party

Slack Third party

Learning resources

GitHub Copilot

Foundry IQ

Building a developer knowledge base
for GitHub Copilot

The answers to your engineering questions
may require many sources

Connect GitHub Copilot to
the GitHub MCP server

1. Add `.vscode/mcp.json`

Connect GitHub Copilot to
a document knowledge base

Connect GitHub Copilot to
a multi-source knowledge base