L2: Tools Tab & Knowledge Management
📋 Agenda
Estimated reading time: ~25 minutes | Hands-on Portal Lab
Learning outcomes:
- ✅ Navigate the Tools Tab and understand its connector ecosystem
- ✅ Configure knowledge sources: file upload, SharePoint native connector, Azure AI Search
- ✅ Connect an MCP server to an agent without writing code
- ✅ Design a knowledge base structure appropriate for your organization
Prerequisites:
- Completed L1 — have a Prompt Agent created and tested
1. Problem Statement
1.1. The Limited Tool Model Is Gone
In Foundry Classic, adding tools meant choosing from a 4-item dropdown:
File Search | Code Interpreter | Function | Bing Search
This was sufficient for prototype work. Production agents in enterprise environments need to connect to:
- SharePoint document libraries that update automatically (no manual upload)
- Internal APIs authenticated via Entra ID
- External data feeds via standardized protocols (MCP)
- Cross-department data in Azure Fabric or SAP
The old dropdown model required developers to build custom connectors for each integration. New Foundry replaces this with the Tools Tab — a centralized management surface with 1,400+ pre-built connectors and native MCP support.
2. What Is the Tools Tab?
2.1. Technical Definition
The Tools Tab (Tab Công cụ) is the centralized management interface in Microsoft Foundry (New) where you discover, configure, and govern all tool integrations available to an agent. It separates tool configuration from agent configuration — tools are reusable across multiple agents in the same project.
2.2. Glossary & Vocabulary
| Term | Vietnamese Meaning & Explanation |
|---|---|
| MCP (Model Context Protocol) | Giao thức chuẩn để kết nối AI model với external tools — do Anthropic khởi xướng, Microsoft tích hợp native vào Foundry |
| Connector | Bộ kết nối sẵn có với một dịch vụ cụ thể (SharePoint, SAP, Salesforce) — thay thế việc tự viết integration code |
| RAG (Retrieval-Augmented Generation) | Kỹ thuật bổ sung context từ tài liệu bên ngoài vào LLM để trả lời dựa trên dữ liệu thực |
| Vector Index | Cơ sở dữ liệu lưu trữ document embeddings để tìm kiếm ngữ nghĩa (semantic search) |
| Chunking | Quá trình chia nhỏ tài liệu thành các đoạn (chunks) trước khi embed |
3. Lab A: Knowledge Management
3.1. Three Knowledge Source Options
New Foundry offers three distinct approaches to connecting knowledge:
| Option | Best For | Update Mechanism |
|---|---|---|
| File Upload | Quick prototyping, static documents | Manual re-upload |
| SharePoint Connector | Org documents that change frequently | Live sync with SharePoint |
| Azure AI Search | Large-scale, production-grade search | Managed indexer pipeline |
3.2. Option 1 — File Upload (Quick Start)
Build → Agents → [Your Agent] → Knowledge section
→ "Add a data source"
→ "Upload files"
→ Drag-and-drop files or click Browse
→ Wait for indexing (progress badge)
Supported formats:
| Format | Use Case | Notes |
|---|---|---|
.pdf | Policies, reports, manuals | Supports scanned PDF via OCR |
.docx | Word documents | Preserves heading structure |
.pptx | Presentation slides | Extracts text from all slides |
.md | Technical documentation | Best format for structured content |
.txt | FAQ, plain text | Fastest to index |
.json | Structured data | Good for key-value FAQ data |
.xlsx, .csv — for tabular data analysis, use Code Interpreter tool instead of File Search. File Search is optimized for prose documents, not spreadsheet data.
Indexing lifecycle:
New Foundry does not maintain document versions. If you upload a new version of a policy document without deleting the old one, both versions co-exist in the index — the agent may retrieve outdated information. Always delete the old file before uploading a revised version.
3.3. Option 2 — SharePoint Connector (Recommended for Production)
This is the most significant improvement over Foundry Classic for enterprise teams:
Build → Agents → [Your Agent] → Knowledge section
→ "Add a data source"
→ "SharePoint"
→ Authenticate with your M365 account
→ Browse: Select SharePoint site → Document Library → Folder
→ Configure sync frequency: Real-time / Daily / Manual
→ Save
Advantage over File Upload:
3.4. Knowledge Base Design Pattern
Naming convention:
| Good | Avoid |
|---|---|
customer-support-policies-2025 | kb1 |
product-catalog-electronics | my-knowledge-base |
hr-employee-handbook-v3 | docs |
4. Lab B: Connecting an MCP Server
Model Context Protocol (MCP) is the biggest new capability in New Foundry's Tools Tab. It allows any agent to connect to any external tool or data source that exposes an MCP-compliant endpoint — without custom integration code.
4.1. What Is MCP?
MCP acts as a universal adapter (bộ chuyển đổi toàn năng): instead of writing a custom function for each external API, you connect to a pre-built MCP server that already handles authentication, error handling, and data formatting.
4.2. Connect an MCP Server
Build → Agents → [Your Agent] → Tools section
→ "+ Add"
→ "Custom" → "Model Context Protocol (MCP)" → "Create"
→ Fill in:
Server Label: outlook-mail
Server URL: https://mcp.ai.azure.com/servers/outlook
Authentication: Microsoft Entra ID
→ Click "Connect"
→ Verify: server appears in Tools list with green status
Microsoft provides hosted MCP servers at mcp.ai.azure.com for common services (Outlook, OneDrive, Azure DevOps, v.v.) — these are pre-authenticated and maintained by Microsoft. For custom internal APIs, your team can host a private MCP server and provide the URL.
4.3. Manage Exposed Tools
Once an MCP server is connected, you can control which tools the agent can call:
Tools Tab → [MCP Server Name] → "Manage tools"
→ Toggle individual tools ON/OFF
→ Example: Outlook MCP exposes:
✅ send_email (enabled)
✅ read_recent_emails (enabled)
❌ delete_email (disabled — safety)
This granular control follows the principle of least privilege (nguyên tắc quyền tối thiểu) — agents only have access to the tools they genuinely need.
5. Capstone Lab: Internal Innovation Assistant
Đây là bài lab tổng hợp sử dụng toàn bộ tính năng của Agent Builder trong Microsoft Foundry (New). Hoàn thành lab này, bạn đã thực hành đầy đủ mọi khái niệm từ L1 và L2.