flywheel-ai 0.8.4

Flywheel CLI and MCP server — business intelligence from the terminal and Claude Code

Flywheel AI

Business intelligence CLI and MCP server for Claude Code.

Flywheel connects your meetings, contacts, pipeline, and business context to Claude Code — so your AI assistant knows your business as well as you do.

Quick Install

curl -fsSL https://flywheel.lumifai.tech/install.sh | bash

Or install with uv tool (recommended — do not use pip install):

uv tool install flywheel-ai
mkdir -p ~/.flywheel/bin
ln -sf ~/.local/bin/flywheel ~/.flywheel/bin/flywheel
export PATH="$HOME/.local/bin:$HOME/.flywheel/bin:$PATH"

Check for a conflicting pip install (common on macOS):

which -a flywheel
python3 -m pip show flywheel-ai flywheel-cli
ls ~/Library/Python/*/bin/flywheel 2>/dev/null

If pip shows a package or which points at Library/Python, remove it:

python3 -m pip uninstall -y flywheel-ai flywheel-cli
rm -f ~/Library/Python/*/bin/flywheel ~/Library/Python/*/bin/flywheel-mcp
hash -r

What You Get

CLI Commands

Command	Description
flywheel login	Authenticate via browser OAuth
flywheel login --headless	Authenticate with email/password credentials
flywheel status	Check login state and tenant info
flywheel focus list	List your focuses (departments/teams)
flywheel focus switch <name>	Switch active focus
flywheel agent setup	Install browser automation deps
flywheel agent start	Start local browser agent
flywheel agent register-deeplink	Register flywheel:// so Prep opens Terminal (macOS)
flywheel setup-claude-code	Register MCP servers with Claude Code

MCP Tools (available in Claude Code)

Once registered, Claude Code gains access to:

• flywheel_run_skill — Run meeting-prep, company-intel, and other skills
• flywheel_read_context — Search your business knowledge base
• flywheel_write_context — Store business intelligence
• flywheel_fetch_meetings — List meetings from your calendar
• flywheel_sync_meetings — Sync meetings from connected providers
• flywheel_list_leads — View your pipeline contacts
• flywheel_draft_lead_message — Draft outreach messages
• And more — run flywheel setup-claude-code to see the full list

Requirements

• macOS (Apple Silicon or Intel)
• Python 3.10+
• Claude Code (for MCP tools)

Configuration

All configuration is handled automatically. Fresh installs write ~/.flywheel/env with FLYWHEEL_API_URL (override via env when running install.sh). Existing values are never overwritten.

Env Variable	Default	Description
FLYWHEEL_API_URL	Hosted backend	Override API endpoint
FLYWHEEL_SUPABASE_URL	Hosted Supabase	Override auth endpoint
FLYWHEEL_SUPABASE_ANON_KEY	Hosted anon key	Supabase anon key for login
FLYWHEEL_CALLBACK_PORT	54321	Local OAuth callback port (http://localhost:<port>/callback)
FLYWHEEL_DIR	~/.flywheel/	Local config directory
FLYWHEEL_LOGIN_EMAIL	unset	Optional headless login email for flywheel login --headless
FLYWHEEL_LOGIN_PASSWORD	unset	Optional headless login password for flywheel login --headless

For local development, backend/.env sets FLYWHEEL_CALLBACK_PORT=54324 to avoid clashing with local Supabase on port 54321.

Override API URL at install time:

FLYWHEEL_API_URL=https://your-backend.example.com \
  curl -fsSL https://flywheel.lumifai.tech/install.sh | bash

flywheel upgrade ensures ~/.flywheel/env exists (if missing) and re-registers the flywheel:// deeplink on macOS.

Publishing & Maintenance

• PyPI package: https://pypi.org/project/flywheel-ai/
• Install script (public gist): https://gist.github.com/Sharan0516/8fb4a3755b3c3e2a2abfae1c949b574a

When you update scripts/install.sh, sync the gist (gist owner):

gh gist edit 8fb4a3755b3c3e2a2abfae1c949b574a scripts/install.sh

The installer calls flywheel agent register-deeplink after login so Prep can open Terminal without a repo checkout. To publish a new CLI version to PyPI:

1. Bump version in cli/pyproject.toml and cli/flywheel_cli/__init__.py (both must match or the publish job will fail)
2. Merge the PR to main
3. The Publish CLI to PyPI workflow auto-runs, tests, and publishes — nothing else needed

If the merge does not include a version bump, the workflow runs but skips the publish step automatically.

One-time repo setup: add a GitHub Actions secret PYPI_API_TOKEN on Lumif-ai/flywheel-v2 scoped to the flywheel-ai PyPI project.

Manual dispatch options (Actions → Publish CLI to PyPI → Run workflow):

• "Publish to Test PyPI" — dry-run against test.pypi.org before a real release
• "Publish even if version unchanged" — force re-upload after a failed publish run

Local publish:

Copy cli/.env.example → cli/.env and set your PyPI token:

cp cli/.env.example cli/.env
# edit cli/.env — UV_PUBLISH_TOKEN=pypi-...
./scripts/publish-cli.sh

Or pass inline without .env:

UV_PUBLISH_TOKEN=pypi-XXXX ./scripts/publish-cli.sh

License

MIT