@basaba/tinyclaw
v0.1.2
Published
A Lobster workflow runner powered by GitHub Copilot — run, schedule, and orchestrate LLM workflows from the terminal
Readme
🦞 TinyClaw
TinyClaw is an automation engine that is meant to run determenistic workflows locally with native integration with LLM/coding agents. Think of it as Power Automate/CI-CD that is running locally and work with full coding agent capabilities. It uses Lobster as it's workflow execution engine.
What is Lobster?
Lobster is OpenClaw's workflow shell — a JSON-first pipeline DSL language for deterministic, resumable automation. Pipelines are composed of typed stages connected by pipes, with built-in support for approval gates, retry logic, state management, and structured input/output.
Note: TinyClaw uses forked version of lobster that has some additions located here.
While Lobster is powerful, but it was built to handle only the execution core only and not the UX around (e.g. the scheduling), this project aimed to bridge this gap, and more improtantly offer more capabilities to make it work with copilot and agency mcps. It aims to make it simple to author and schedule execution pipelines locally, that otherwise would require the dev to write custom scripts and run it with cronjobs.
What it does
- LLM-powered workflows — Use
llm.invoke --provider copilotin any Lobster pipeline to call copilot for reasoning, summarization, code review, or structured data extraction. - Scheduler & TUI — Schedule workflows on cron-like intervals and manage them from a terminal UI. A background daemon handles execution, approval gates, and run history.
- MCP integration — Call external tools via the MCP. TinyClaw discovers and connects to MCP servers defined in your config, with native support for agency MCP servers (mail, Teams, and more) out of the box.
- Built-in commands — Send Teams messages, search and compose emails, monitor Azure DevOps PRs — all from within workflows.
- Plugin system — Drop
.jsfiles in the plugin directory to add custom commands without modifying source.
Short examples
Monitor ADO PRs given critirea and send as teams message.
steps:
- id: monitor
pipeline: >
ado.pr.monitor
--org "${org}"
--project "${project}"
--repository "${repository}"
--creator "${creator}"
--days "${days}"
| where changed==true # Remove already processed PRs
| copilot --prompt "Summarize these PRs"
| teams.send --self # send to self via Teams agency MCPSummarize mails and send to self.
steps:
- id: mail-track
pipeline: >
mail.search --folder inbox --top 5 --order-by newest
| copilot --prompt "Summarize these mails"
| teams.send --selfSee more:
- Language Specification — Full reference for Lobster syntax, cli doc, built-in commands, and TinyClaw extensions.
- Examples — Ready-to-run workflows for mail, Teams, ADO PR monitoring, and more.
Installation
From npm
npm install -g tinyclawFrom source
git clone https://github.com/basaba/tinyclaw.git
cd tinyclaw
npm install
npm run build
npm linkQuick Start
Launch the TUI:
tinyclawRun a workflow file:
tinyclaw examples/approval-demo.yamlRun a pipeline string:
tinyclaw -p "llm.invoke --provider copilot --prompt 'Hello'"License
See LICENSE for details.
