mcp-integration

Help a learner give an agent tools through the Model Context Protocol (MCP) — both client-side MCP and Foundry Hosted MCP tools. Use when connecting an agent to an MCP server (e.g. the GitHub MCP server), adding a remote tool, or reasoning about MCP approval/governance.

Jump to install

Repository: microsoft/Building-AI-Agents-From-Zero-To-Production
Last source activity: July 13, 2026 at 12:47
Detected SKILL.md language: English
Stars: 177
Forks: 70

Install options

The review-first prompt is selected by default. You can switch to a direct command or download a local copy.

Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.

A direct command skips the review prompt. Inspect the source before running it.

npx skills add https://github.com/microsoft/Building-AI-Agents-From-Zero-To-Production --skill mcp-integration

The command stays on one line. Scroll horizontally to inspect it before copying.

Prefer a local copy? Download the files currently available to SkillsMP.

Review the source files

Read SKILL.md and any companion files shown by SkillsMP before deciding whether to install.

SKILL.md

Source instructions · Read-only preview

Back to search Back to installation

agent-evaluation

Help a learner evaluate and observe an AI agent on Microsoft Foundry — quality/safety evaluators, tracing, and reading results. Use for Lesson 3 tasks: measuring agent performance, adding observability, or interpreting evaluation output.

2026-07-13

foundry-agent-basics

Help a learner build their first AI agent with the Microsoft Agent Framework (MAF) on Microsoft Foundry. Use when the task is creating a basic agent, wiring up FoundryChatClient, choosing a model, running the DevUI, or understanding lesson 1–2 concepts (instructions, tools, threads).

2026-07-13

hosted-agent-deployment

Help a learner deploy an agent to production as a Microsoft Foundry Hosted Agent, and decide when Capability Hosts are needed. Use for Lessons 4–5: hosted deployment, thread/memory persistence, bring-your-own storage, and enterprise governance.

2026-07-13

View all from this repository

name	mcp-integration
description	Help a learner give an agent tools through the Model Context Protocol (MCP) — both client-side MCP and Foundry Hosted MCP tools. Use when connecting an agent to an MCP server (e.g. the GitHub MCP server), adding a remote tool, or reasoning about MCP approval/governance.

MCP integration

Guidance for adding tools to an agent via the Model Context Protocol (MCP).

Guardrails (always apply)

Terminology: MCP, Hosted MCP tools (Foundry-managed), Microsoft Learn MCP server.
Client-side MCP uses MCPStreamableHTTPTool from the current agent-framework surface. The old top-level HostedMCPTool symbol is gone — hosted tools are configured through the Foundry hosted-tool types.
Secrets (PATs, tokens) come from .env only. In this course the GitHub MCP server is passed a Bearer PAT via headers read from GITHUB_PERSONAL_ACCESS_TOKEN — never hardcode it.

Concepts a learner should grasp

Why MCP: one open protocol lets an agent discover and call tools hosted anywhere, instead of bespoke per-tool integrations.
Client MCP vs Hosted MCP: client-side, your process connects to an MCP endpoint and executes tool calls; Hosted MCP runs the connection inside Foundry with managed identity, approvals and observability.
Approval workflows: hosted MCP tools can require human approval before a tool runs — important for governance in production (covered in Lesson 5).

Worked example in this repo

lesson-2-agent-development/task-recommendation-agent.py connects an agent to the remote GitHub MCP server as a tool (Lesson 1 Scenario 2), passing a Bearer PAT via headers. Use it as the reference implementation.

How to help

Start from task-recommendation-agent.py and adapt the MCP endpoint / headers.
For hosted/enterprise scenarios, direct the learner to lesson-5-hosted-agents-production/ for the approval-workflow and governance discussion.
Confirm the target MCP server URL and required auth before writing code.

Validate

python -m py_compile <file>.py.
The agent should list/call at least one MCP tool and return a grounded answer.

References

lesson-2-agent-development/README.md (sample catalog), lesson-5-hosted-agents-production/README.md

mcp-integration

Source facts

Install options

Review the source files

More from this repository

More from this repository

MCP integration

Guardrails (always apply)

Concepts a learner should grasp

Worked example in this repo

How to help

Validate

References

MCP integration

Guardrails (always apply)

Concepts a learner should grasp

Worked example in this repo

How to help

Validate

References