Grafana Assistant CLI

[!NOTE] This package is currently in public preview. APIs and features may change without notice.

CLI for interacting with Grafana Assistant via the A2A (Agent-to-Agent) API.

Installation

uvx / pip (Recommended)

Run directly without installing:

uvx grafana-assistant

Or install globally:

pip install grafana-assistant

Homebrew (macOS / Linux)

brew install grafana/grafana/grafana-assistant

Or, if you prefer to tap first:

brew tap grafana/grafana
brew install grafana-assistant

Manual download

Download the latest pre-built binary for your platform from the latest release page, extract it, and add the grafana-assistant binary to a directory in your PATH.

On Windows, download the .zip archive instead and add grafana-assistant.exe to your PATH.

Quick Start

# Add your Grafana instance
grafana-assistant config set-instance mystack \
  --url https://mystack.grafana.net

# Set it as the default
grafana-assistant config use-instance mystack

# Authenticate via browser
grafana-assistant auth

# Start interactive chat
grafana-assistant chat

The auth command opens your browser for a one-time login. After that, the CLI manages short-lived tokens that auto-refresh — no service account needed.

[!IMPORTANT] The Assistant CLI User role is required to obtain access tokens via grafana-assistant auth. This role is automatically granted to users with the Editor role or above. For custom roles, the grafana-assistant-app.tokens:access permission must be included.

Configuration is stored in ~/.config/grafana-assistant/config.yaml by default, or ./grafana-assistant.yaml for project-specific configs. Session state (last context ID for --continue) is stored in ~/.config/grafana-assistant/state.yaml.

For detailed setup instructions, see docs/SETUP.md. For the full config file reference, see docs/CONFIGURATION.md.

Usage

For detailed usage guides, see Setup & Auth, Chat & Prompt, and Tunnel.

Interactive Chat (Recommended)

Start a full-screen interactive chat session with real-time streaming responses and conversation continuation.

# Start interactive chat
grafana-assistant chat

# With a specific agent
grafana-assistant chat --agent investigation_agent

# Resume a previous conversation by context ID
grafana-assistant chat --context "previous-context-id"

# Continue the last chat session
grafana-assistant chat --continue

# Using a specific instance
grafana-assistant chat --instance prod

Commands:

• /clear or /new - Start a new conversation
• /exit, /quit, or /q - Exit
• /help - Show help

Keyboard Shortcuts:

• Enter - Send message
• Ctrl+C - Cancel current request or exit
• Ctrl+D - Exit

Single-Shot Prompt (for Scripting)

Send a single message and receive the response. Useful for automation and scripting.

# Send a single prompt
grafana-assistant prompt "Analyze my dashboard"

# Using a specific instance from config
grafana-assistant prompt "Analyze my dashboard" --instance prod

# Using explicit URL and token
grafana-assistant prompt "Analyze my dashboard" \
  --url https://your-stack.grafana.net \
  --token $GRAFANA_SA_TOKEN

# Set a custom timeout (default: 300 seconds)
grafana-assistant prompt "Query error rates from Prometheus" --timeout 600

# Continue a conversation with context ID
grafana-assistant prompt "Now show me the last 24 hours" --context "previous-context-id"

# Continue the most recent chat/prompt session
grafana-assistant prompt "Now show me the last 24 hours" --continue

# Output as JSON (for scripting)
grafana-assistant prompt "Analyze my dashboard" --json

# Generate an AGENTS.md file for the current directory
grafana-assistant agents-md

CLI Reference

Chat Command (Interactive)

Option	Description	Default
--url, -u	Grafana instance URL	-
--instance, -i	Instance name from config	-
--context, -c	Context ID for conversation continuation	-
--continue	Continue the previous chat session	false
--timeout	Timeout in seconds per request	300

Prompt Command (Non-Interactive)

Option	Description	Default
--url, -u	Grafana instance URL	-
--instance, -i	Instance name from config	-
--wait, -w	Wait for completion	true
--timeout	Timeout in seconds	300
--context, -c	Context ID for conversation threading	auto-generated
--continue	Continue the previous chat session	false
--json	Output as JSON	false

Auth Command

Command	Description
auth	Authenticate via browser
auth --instance <name>	Authenticate a specific instance

Agent MD Command

Generate an AGENTS.md file with practical guidance for using grafana-assistant in a project. When run interactively it reads configured instances and lets you choose which ones to include with their real URLs pre-filled.

# Generate AGENTS.md in the current directory (interactive)
grafana-assistant agents-md

# Generate into a specific directory and overwrite existing files
grafana-assistant agents-md ~/projects/my-app --force

# Skip interactive prompts (CI / scripting)
grafana-assistant agents-md --non-interactive

Option	Description	Default
--output, -o	Output filename to create in target directory	AGENTS.md
--force, -f	Overwrite output file if it already exists	false
--dry-run	Print generated content to stdout instead of writing a file	false
--non-interactive	Skip interactive prompts (use generic placeholders)	false

Config Commands

Command	Description
config set-instance <name>	Add or update an instance
config use-instance <name>	Set the current instance
config list	List all configured instances
config current	Show the current instance
config delete-instance <name>	Remove an instance
config add-project <name> <path>	Add a project directory
config list-projects	List all configured projects
config remove-project <name>	Remove a project

Tunnel Command

Run a tunnel that allows Grafana Assistant to execute tools on your machine.

# Authenticate (includes tunnel access)
grafana-assistant auth

# Add a project for the assistant to access
grafana-assistant config add-project my-app ~/projects/my-app

# Start tunnel with filesystem access (default)
grafana-assistant tunnel connect

# Start tunnel with terminal access (use with caution)
grafana-assistant tunnel connect --terminal

The auth command includes the tunnel:connect scope, so no separate tunnel auth step is needed.

The tunnel provides:

• Filesystem tool: Read-only access to local files with project-based configuration
  • Configure named projects (e.g., my-app → ~/projects/my-app)
  • Access files with project-relative paths (e.g., src/main.go)
  • Search code with grep and find_path actions
  • Add/remove projects dynamically mid-session
• Terminal tool: Execute shell commands (with configurable allow/deny lists)

For security, the tunnel:

• Blocks access to sensitive files (.ssh, .env, private keys, etc.) by default
• Blocks dangerous commands (rm -rf /, mkfs, fork bombs, etc.) by default
• Runs commands with a minimal environment (only PATH, HOME, USER, etc.)
• Requires explicit project configuration before accessing codebases

See docs/TUNNEL.md for the full tunnel usage guide and docs/CONFIGURATION.md for tunnel configuration options.

Environment Variables

Variable	Description
GRAFANA_URL	Grafana instance URL
GRAFANA_ASSISTANT_CONFIG	Override config file path

JSON Output

When using --json with the prompt command, the output format is:

Prompt Response

{
  "taskId": "a2a-task-123",
  "contextId": "uuid-for-threading",
  "status": "completed",
  "response": "The agent's response text..."
}

Possible status values: completed, failed, timeout, canceled, unknown

Tool Approval Flow

When the Grafana Assistant needs to execute certain tools that require user confirmation (such as running queries or modifying resources), the CLI will prompt you for approval.

How It Works

1. During a chat session, when a tool requires approval, you'll see a prompt:

Approve execute_query - Execute a database query? [y]/[n]

2. Press y or Y to approve and allow the tool to execute
3. Press n, N, or Esc to deny and skip the tool execution

Approval Keys

Key	Action
y or Y	Approve the tool execution
n or N	Deny the tool execution
Esc	Deny the tool execution

When you deny a tool, the assistant will be informed and will continue the conversation without executing that tool.

A2A Protocol

This CLI uses the A2A (Agent-to-Agent) Protocol v0.3.0 for communication:

• Streaming: Real-time SSE streaming of agent responses
• Context threading: Continue conversations with context IDs
• Standard protocol: JSON-RPC 2.0 over HTTP with SSE

For more details, see the A2A Protocol Specification.

Prerequisites

1. A Grafana instance with the Assistant plugin enabled
2. A Grafana user account for browser-based authentication

Contributing

See CONTRIBUTING.md for development setup, testing, and snapshot test instructions.

License

This software is licensed under the Grafana Enterprise Plugin License Agreement.