CLI flags
| Command | Description |
|---|
auggie | Start Auggie in interactive mode |
auggie --print "Summarize the staged changes" (-p) | Run one instruction in print mode and exit |
auggie --print --quiet "Summarize the staged changes" | Show only the final assistant message |
auggie --print --compact "Summarize the staged changes" | Show compact streaming output in print mode |
auggie --print --show-credits "Summarize the staged changes" | Include a credit usage summary at the end of the run |
auggie --print --output-format json "Summarize the staged changes" | Output the response in structured JSON format for automation workflows |
| Command | Description |
|---|
auggie "Fix the typescript errors" | Provide an initial instruction in interactive mode |
auggie --print "Summarize the staged changes" | Provide an instruction and exit |
cat file | auggie --print "Summarize this data" | Pipe content through stdin |
auggie --print "Summarize this data" < file.txt | Provide input from a file |
auggie --instruction "Fix the errors" | Provide an initial instruction in interactive mode |
auggie --instruction-file /path/to/file.txt | Provide the initial instruction from a file |
auggie --print --queue "Write tests" --queue "Summarize remaining risks" "Fix the bug" | Queue additional instructions for sequential execution |
auggie --image screenshot.png "Explain this UI error" | Attach one or more images to the initial prompt |
auggie --print --enhance-prompt "Fix the flaky test" | Run the prompt enhancer before sending the instruction |
Custom Commands
| Command | Description |
|---|
auggie command list | List available custom commands discovered from command directories |
auggie command <command-name> | Execute a custom command from .augment/commands/ or ~/.augment/commands/ |
auggie command help | Show help for custom commands and command discovery |
~/.augment/commands/<name>.md - Global commands (user-wide)./.augment/commands/<name>.md - Project commands (workspace-specific)~/.claude/commands/<name>.md - Claude Code user commands./.claude/commands/<name>.md - Claude Code workspace commands
Commands are resolved in order of precedence, with Auggie-specific locations taking priority over Claude Code locations.
Examples:
# Execute a custom deployment command
auggie command deploy-staging
# Execute a code review command
auggie command security-review
# List available commands
auggie command list
Sessions
| Command | Description |
|---|
auggie --continue (-c) | Resume the most recent conversation |
auggie --resume (-r) | Interactively choose a saved session to resume |
auggie --resume <sessionId> | Resume a specific session by ID or prefix |
auggie --dont-save-session | Do not save the conversation to the local history |
auggie session list | List saved sessions from the current workspace |
auggie session delete | Interactively delete saved sessions |
auggie session delete --all | Delete saved sessions from all workspaces |
auggie session share [sessionId] | Generate a shareable link for a saved session |
Configuration
| Command | Description |
|---|
auggie --workspace-root /path/to/project | Specify the root of the workspace |
auggie --add-workspace /path/to/another/project | Index an additional workspace alongside the primary workspace |
auggie --rules /path/to/rules.md | Additional rules to append to workspace guidelines |
auggie --model "name" | Select the model to use (accepts long or short names from the model list) |
auggie --persona "persona-id" | Select the agent persona to use for this session |
auggie --augment-cache-dir /path/to/cache | Override the Augment cache directory |
auggie --retry-timeout 30 | Set the rate-limit retry timeout in seconds |
auggie --print --max-turns 10 "Review this diff" | Limit the number of agentic turns in print mode |
auggie --allow-indexing | Skip the interactive indexing confirmation screen |
Skills are loaded automatically from
.augment/skills/ and
.claude/skills/ directories in both your workspace and home directory. See
Skills for more information.
Models
List out available models and their short names to be passed into the --model flag
| Command | Description |
|---|
auggie models list | List available models |
auggie models list --json | Output full model metadata as JSON for downstream tooling |
| Command | Description |
|---|
auggie --mcp-config /path/to/mcp.json | Load MCP server configuration from a JSON file for the current run |
auggie --permission launch-process:allow | Set a tool permission rule for the current run |
auggie --remove-tool <tool-name> | Remove a specific tool by name for the current session. Can be specified multiple times. |
auggie --shell zsh | Override the shell used by process tools for the current session |
auggie --startup-script "source .venv/bin/activate" | Run a startup script before each shell command |
auggie --startup-script-file /path/to/script.sh | Load a startup script from a file |
auggie tools list | List all available tools and their current status |
auggie tools remove <tool-name> | Persistently remove a tool by adding it to the removedTools list in settings.json |
auggie tools add <tool-name> | Re-enable a previously removed tool by removing it from the removedTools list |
auggie tools schemas | Print the tool definitions and input schemas |
# Disable the web-fetch tool for this session
auggie --remove-tool web-fetch
# Disable multiple tools for this session
auggie --remove-tool web-fetch --remove-tool web-search
# Persistently disable a tool
auggie tools remove launch-process
# Re-enable a previously disabled tool
auggie tools add launch-process
# See all tools and their status
auggie tools list
Command-line
--remove-tool flags take precedence over settings. For fine-grained control over tool behavior (allow, deny, ask-user), see
Permissions.
Plugins and Marketplaces
| Command | Description |
|---|
auggie plugin marketplace add <github-repo> | Add a plugin marketplace from a GitHub repository |
auggie plugin marketplace list | List all installed plugin marketplaces |
auggie plugin marketplace update [name] | Update a specific marketplace or all marketplaces |
auggie plugin marketplace remove <name> | Remove an installed marketplace |
auggie plugin list | List all available plugins from installed marketplaces |
auggie plugin install <plugin-id> | Enable a plugin |
auggie plugin install <plugin-id> --disable | Disable a plugin |
Plugin commands are available when the plugin marketplace feature is enabled for your account.
MCP and integrations
| Command | Description |
|---|
auggie mcp add [options] <name> | Create or update a named MCP server configuration |
auggie mcp add-json <name> <json> | Add an MCP server from JSON configuration |
auggie mcp list | Display all configured MCP servers |
auggie mcp remove <name> | Remove a named MCP server configuration |
| Command | Description |
|---|
auggie --mcp-config {key: value} | MCP configuration as a JSON string |
auggie --mcp-config /path/to/mcp.json | MCP configuration from a JSON file |
You can define MCP servers persistently in the settings files: ~/.augment/settings.json. Any --mcp-config flags are applied last and override settings.
MCP Server Mode
Run Auggie as an MCP server to expose the codebase-retrieval tool to external AI tools like Claude Code, Cursor, and others.
| Flag | Description |
|---|
--mcp | Run Auggie as an MCP tool server. Uses the current working directory as the workspace by default. |
--mcp-auto-workspace | Enable automatic workspace discovery based on client requests (added in v0.14.0) |
-w /path/to/project | Specify a workspace to index |
Automatic Workspace Discovery
The --mcp-auto-workspace flag enables dynamic workspace discovery in MCP mode. When enabled:
- The
codebase-retrieval tool accepts a directory_path parameter to specify which workspace to search
- Workspaces are indexed on-demand when first accessed
- Multiple workspaces can be searched within a single MCP server session
This is useful when the MCP client (e.g., Claude Code) needs to work with multiple projects or when the workspace isn’t known at startup time.
You can combine --mcp-auto-workspace with -w to pre-index a primary workspace at startup while still allowing dynamic discovery of additional workspaces. This is useful for large workspaces that take time to index, or to reduce latency on the first query to your main project.
Examples:
# MCP server with auto-discovery (recommended)
auggie --mcp --mcp-auto-workspace
# Pre-index a workspace, allow dynamic discovery of others
auggie --mcp --mcp-auto-workspace -w /path/to/primary/project
# Use only a single workspace path
auggie --mcp -w /path/to/project
# Use current working directory as the workspace
auggie --mcp
When using --mcp-auto-workspace, the first query to a new workspace may take longer as the workspace is indexed. Subsequent queries to the same workspace will be fast.
Authentication
| Command | Description |
|---|
auggie login | Login to Augment and store the token locally |
auggie logout | Remove the locally stored token |
auggie token print | Print the locally stored session JSON for automation |
auggie token revoke | Revoke all authentication tokens for the current user |
Additional commands
| Command | Description |
|---|
auggie account status | Show account and billing information |
auggie upgrade [--skip-confirmation] | Upgrade Auggie to the latest version |
auggie --help | Show help |
auggie --version | Show version |
Environment Variables
| Variable | Description |
|---|
AUGMENT_SESSION_AUTH | Authentication JSON. |
GITHUB_API_TOKEN | GitHub API token |
AUGMENT_DISABLE_AUTO_UPDATE | Disable automatic CLI updates |
Shell Environment
When Auggie executes shell commands using the launch-process tool, it sets the following environment variable:
| Variable | Description |
|---|
AUGMENT_AGENT | Set to 1 when a command is executed by Auggie. Scripts can check for this variable to detect if they are being run by the agent. |
if [ -n "$AUGMENT_AGENT" ]; then
echo "Running inside Auggie"
# Adjust behavior for agent execution
fi
See Also
