ensemble-mcp docs
Presentation GitHub

CLI Reference

Complete reference for all ensemble-mcp CLI commands, flags, and options.

Entry Point

ensemble-mcp [command] [options]

The CLI is registered as ensemble-mcp via the pyproject.toml entry point (ensemble_mcp.__main__:main).

Running ensemble-mcp with no command defaults to serve.


Commands

serve (default)

Start the MCP server on stdio. This is the default when no subcommand is given.

ensemble-mcp
ensemble-mcp serve

The server:


web

Start the local web dashboard.

ensemble-mcp web [options]
FlagTypeDefaultDescription
--portint8787Port to bind on
--no-openflagfalseDon't auto-open browser
--reports-dirpathautoDirectory containing Bug Hunter report files

Examples:

# Start dashboard on default port
ensemble-mcp web

# Custom port, no auto-open
ensemble-mcp web --port 9000 --no-open

# Specify reports directory
ensemble-mcp web --reports-dir ./reports

The dashboard binds to 127.0.0.1 only (local access, no authentication required). If the port is already in use, an error message suggests the next port.

Reports directory auto-detection: When --reports-dir is not specified, the dashboard looks for a reports/ directory in (1) the current working directory, then (2) the git repository root.


install

Auto-detect AI tools and register ensemble-mcp in their MCP configs.

ensemble-mcp install [options]
FlagTypeDefaultDescription
--localflagfalseRegister in project-local configs instead of global
--project-pathpathcwdProject root directory
--toolsstringallComma-separated list of tools (e.g. --tools opencode,cursor)
--dry-runflagfalseShow plan without making changes
--yes, -yflagfalseSkip confirmation prompt

Supported tool names: opencode, claude_code, copilot, cursor, windsurf, devin

Examples:

# Auto-detect and register globally
ensemble-mcp install

# Register only in Cursor and OpenCode
ensemble-mcp install --tools cursor,opencode

# Project-local registration
ensemble-mcp install --local

# Preview without changes
ensemble-mcp install --dry-run

# Non-interactive (CI/CD)
ensemble-mcp install --yes

The installer:

  1. Detects installed AI tools by checking for config directories
  2. Displays a plan showing what will be registered
  3. Creates backups of existing config files (.bak)
  4. Registers the MCP server entry
  5. Copies bundled agent files to tool-specific directories
  6. Copies bundled skill files to project-local directories

uninstall

Remove ensemble-mcp registration from AI tool configs.

ensemble-mcp uninstall [options]
FlagTypeDefaultDescription
--localflagfalseRemove from project-local configs
--project-pathpathcwdProject root directory
--toolsstringallComma-separated list of tools
--remove-agentsflagfalseAlso remove agent/skill files
--clean-dataflagfalseRemove ~/.cache/ensemble-mcp/ and ~/.config/ensemble-mcp/
--dry-runflagfalseShow plan without making changes
--yes, -yflagfalseSkip confirmation prompt

Examples:

# Remove from all detected tools
ensemble-mcp uninstall

# Full cleanup: deregister, remove agents, delete data
ensemble-mcp uninstall --remove-agents --clean-data --yes

# Preview uninstall plan
ensemble-mcp uninstall --dry-run

add-agents

Copy bundled agent files to tool-specific directories without MCP registration.

ensemble-mcp add-agents [options]
FlagTypeDefaultDescription
--localflagfalseCopy to project-local agent dirs
--project-pathpathcwdProject root directory
--toolsstringallComma-separated list of tools
--dry-runflagfalseShow plan without making changes
--yes, -yflagfalseSkip confirmation prompt

Bundled agents include the 7-agent orchestration pipeline: team-ensemble, team-scope, team-craft, team-forge, team-trace, team-lens, team-signal.

Examples:

# Copy agents globally for all detected tools
ensemble-mcp add-agents

# Copy agents locally for OpenCode only
ensemble-mcp add-agents --local --tools opencode

add-skills

Copy bundled skill files to tool-specific directories without MCP registration.

ensemble-mcp add-skills [options]
FlagTypeDefaultDescription
--localflagdefaultCopy to project-local skill dirs (default)
--globalflagfalseCopy to global skill dirs instead
--project-pathpathcwdProject root directory
--toolsstringallComma-separated list of tools
--dry-runflagfalseShow plan without making changes
--yes, -yflagfalseSkip confirmation prompt
Note

add-skills defaults to local scope (unlike add-agents which defaults to global). Use --global to override.

Bundled skills include the ensemble-mcp-workflow skill that teaches AI agents when and how to invoke ensemble-mcp tools.

Examples:

# Copy skills locally (default)
ensemble-mcp add-skills

# Copy skills globally
ensemble-mcp add-skills --global

# Preview what would be copied
ensemble-mcp add-skills --dry-run

Environment Variables

All settings fields can be overridden via environment variables with the ENSEMBLE_MCP_ prefix:

VariableTypeDescription
ENSEMBLE_MCP_DB_PATHpathOverride database file location
ENSEMBLE_MCP_CACHE_DIRpathOverride cache directory
ENSEMBLE_MCP_MODEL_DIRpathOverride model directory
ENSEMBLE_MCP_MAX_PATTERNSintMaximum stored patterns
ENSEMBLE_MCP_DEFAULT_TOP_KintDefault search result count
ENSEMBLE_MCP_DEFAULT_MIN_SCOREfloatMinimum similarity threshold
ENSEMBLE_MCP_DRIFT_THRESHOLD_ALIGNEDfloatDrift score below which changes are “aligned”
ENSEMBLE_MCP_DRIFT_THRESHOLD_MINORfloatDrift score below which changes are “minor”
ENSEMBLE_MCP_IDEMPOTENCY_KEY_TTL_HOURSintHours before idempotency keys expire

See Configuration for the full list of settings.

Exit Codes

CodeMeaning
0Success
1Error (invalid args, port in use, unknown tool names, etc.)

Next Steps