Implement tool use and environment interaction Study Guide — GH-600

Implement Tool Use and Environment Interaction

This domain tests your ability to configure agent tools, set up MCP servers and registries, scope agents to repositories and branches, enable autonomous code actions, and build safe execution paths with error handling, escalation, and traceability. It accounts for 23% of the exam — the largest single domain — so mastery here is essential for a passing score.

Identifying and Configuring Required Agent Tools and Tool Permissions

Agent tools are discrete capabilities granted to a Copilot agent that allow it to read data, execute commands, or interact with external systems. Tools are declared explicitly; an agent cannot use a capability it has not been granted.

Tool Categories

Permission Scoping

Each tool requires explicit permission to be active. Permissions are scoped at two levels:

• Agent-level: declared in the agent manifest or copilot-instructions.md context block
• Organization-level: administrators enable or disable tool categories for all agents in the org via GitHub Copilot policy settings

The principle of least privilege applies: grant only the tools required for the agent's stated task. Over-provisioning tools is flagged as a policy violation in enterprise audits.

Common Traps

• Assuming tools are inherited from the user's personal Copilot subscription — they are not; tools must be declared per agent.
• Confusing tool permissions (what the agent can call) with repository permissions (what the agent's token can access) — both must be configured independently.
• Believing that adding a tool to an agent manifest automatically grants it write access to GitHub resources — API tools still require a token with appropriate scopes.

Configuring MCP Servers: Adding an MCP Server as a Tool, GitHub Remote MCP Server

Model Context Protocol (MCP) is an open standard that defines how AI agents discover and invoke external tools through a structured server-client interface. An MCP server exposes one or more tools; the agent acts as the MCP client.

Adding an MCP Server as a Tool

MCP servers are configured in the agent's settings file (typically .github/copilot/mcp.json or the equivalent VS Code settings.json block):

{
  "mcpServers": {
    "my-tool-server": {
      "url": "https://my-mcp-server.example.com",
      "transport": "http"
    }
  }
}

Key configuration fields:

GitHub Remote MCP Server

The GitHub Remote MCP Server is a managed MCP server hosted and operated by GitHub. It exposes GitHub-native capabilities (issue read/write, PR management, code search) as MCP-compliant tools without requiring the agent administrator to run any server infrastructure.

The GitHub Remote MCP Server is referenced with the reserved key github in mcp.json:

{
  "mcpServers": {
    "github": {
      "url": "https://api.github.com/mcp",
      "transport": "http"
    }
  }
}

Common Traps

• Using stdio transport for a remote server — stdio is only valid for locally-spawned processes; remote servers require http.
• Thinking the GitHub Remote MCP Server requires a separately provisioned resource — it is accessed via the standard GitHub API endpoint.
• Confusing the MCP server url with the GitHub API base URL — the MCP endpoint path is /mcp, not /v3 or /graphql.