Table of Contents

Tool reference

Use this catalog to configure custom agents or understand tool calls. Check Tools and Skills for the selected conversation's available tools.

All tools remain subject to provider/model support, enablement, active-agent restrictions, and permissions.

Inspect the workspace

Tool Purpose
search_files Find files by name, path, or glob pattern.
read_file Read text, with pagination for long files.
read_image Load a supported image for a model that can process image input.
grep Search file contents with regular expressions.
list_directory List a directory's contents.
current_document Identify the active Visual Studio document.
get_projects_in_solution List projects in the open solution.
get_files_in_project List a project's files, optionally filtered.
get_errors Retrieve compilation errors and warnings, optionally filtered by file and severity.
get_tests Discover tests and filter the results.

Inspection tools work in Agent and Plan when dependencies are available. IDE results require a connected Visual Studio instance and relevant solution state.

Change and validate code

Tool Purpose and availability
write_file Create or edit files. Agent mode.
apply_patch Apply a batch of file additions, updates, or deletions with prevalidation and rollback handling. Agent mode.
move_files Move or rename files. Agent mode.
remove_file Delete one workspace file and update supported loaded project membership. Agent mode.
run_build Build the solution or supported target; available in Agent and Plan.
run_tests Run tests with supported selection/filtering; available in Agent and Plan.
run_cli Execute managed CLI requests and bounded batches. Plan accepts recognized reads only.
shell_command Execute Windows PowerShell 5.1 scripts or recognized managed commands. Plan accepts recognized managed reads, not ordinary scripts.

remove_file does not delete directories or stage Git changes. It rejects unsupported project items and requires unsaved documents/projects to be saved first. Recorded workspace changes support review and undo, subject to conflict checks. Build and test execution can create artifacts even in Plan.

Managed command integrations classify supported Git, GitHub, text, and .NET operations. Configure their rights under Run CLI commands > Integrations. Historical identifiers git_read, git_write, github_read, and github_write are internal compatibility/permission categories, not tools to select for a new agent. Use run_cli with explicit integration rights.

Plan, clarify, and customize

Tool Purpose and availability
update_plan Create or update the active Markdown task plan; Agent and Plan.
read_plan Read the active plan; Agent and Plan.
clarify_requirements Present a structured question with selectable answers; requires an interactive client.
skill Load a skill's instructions for extension-managed providers. Native skill providers use their own integration.
create_skill Create a validated workspace or global skill; Agent mode.
create_agent_profile Create a validated workspace or global .agent.md definition; Agent mode.
ollama_load_model Load a model into an Ollama server's memory; Ollama only.
ollama_unload_model Unload a model from an Ollama server's memory; Ollama only.

Provider-native and MCP tools

Provider-native capabilities are separate from this catalog. Examples include Claude Code's WebSearch/WebFetch, Codex's web.run and image generation, Kimi's search/fetch/skill tools, and Kiro's retained native tools. OpenRouter and DeepSeek can expose native web capabilities; Databricks search is controlled by its provider setting.

These capabilities depend on the model, endpoint, account, and installed CLI. Some cannot be individually disabled. See Providers and tool selection.

Connected MCP servers add their enabled tools dynamically. Custom-agent definitions refer to them as server-id/tool-name or the deliberate wildcard server-id/*.