lola-skill-cli

LOLA market CLI. One binary (lola), one subcommand per skill.

lola login                            # device flow for ALL skills (one approval click)
lola login --skill manage-rentals     # device flow for one skill
lola whoami [--skill manage-rentals]
lola logout [--skill … | --all]

lola rentals list
lola rentals show <id>
lola rentals events <id>
lola rentals advance <id> <event> [--lat … --lng … | --file-ids a,b,c]
lola rentals earnings
lola rentals reviews

Install (dev)

cd cli
npm install
npm run build
npm link            # exposes `lola` on $PATH

Or run without linking:

node dist/bin.js login

Using the MCP server (lola-mcp)

The same operations are exposed as MCP tools over stdio by the lola-mcp binary (installed alongside lola by the same package). Point any MCP host at it — Claude Desktop, Claude Code, Cursor, opencode, Continue, etc.

{
  "mcpServers": {
    "lola": { "command": "lola-mcp" }
  }
}

First-time setup is still the CLI. lola-mcp reads the same ~/.config/lola/credentials.json the CLI writes — there is no login flow inside the MCP server. Order of operations:

1. npm i -g @lola-market-team/skill-cli (puts both lola and lola-mcp on PATH)
2. lola login (one-time, opens a browser for device-flow approval)
3. Add the JSON snippet above to your MCP host and restart it
4. The agent can call lola_status first to confirm the user is paired — it's a no-network local read and won't throw before login

If a tool call fires before login, the error string tells the user exactly what to run.

Where credentials live

~/.config/lola/credentials.json — file mode 0600, one record per skill keyed by client_id. Delete or lola logout to drop a session.

Base URL

Defaults to https://staging.lola.market. Override with the LOLA_BASE_URL environment variable.

Backend wire contract

Each skill's HTTP contract is documented in ../docs/skills/<slug>/skill.yaml. The CLI is a thin wrapper.

Driving the CLI from an LLM / agent

Designed so a shell-spawning agent can use it without prior knowledge. Self-introspection paths:

1. lola --help → top-level workflow, env vars, examples
2. lola <skill> --help → what the skill does, deny-list, error codes, and copy-paste examples
3. lola <skill> <cmd> --help → exact flags and their purpose
4. For state-machine skills: always lola <skill> events <id> before advance — the proxy applies the deny-list, so what comes back is safe to fire. Don't enumerate FSM events from memory.

Output today is JSON for everything except rentals list (small table). Pipe through jq for parsing. A --json global flag for the list commands is on the roadmap.

Errors land as a JSON body with errors[0].extensions.code. Stable codes worth pattern-matching:

| Code | Action | |--------------------------|--------------------------------------------------| | NOT_FOUND | resource doesn't exist OR caller isn't on it | | EVENT_BLOCKED_FOR_SKILL| money/insurance event; deny-list, not a bug | | WRONG_SKILL | token is for another skill; re-login --skill X | | UNAUTHORIZED | session expired; CLI auto-refreshes once | | unknown_event | FSM rejected (wrong state/role) |

Repo plan

Lives under cli/ in lola-market-backend for now while contracts are still moving. Will be extracted to its own repo (lola-cli) once the surface is stable, with:

git filter-repo --path cli/ --to-subdirectory-filter .