FASTRACK

Plain English in the terminal. Workflows out.

FASTRACK is a Node.js CLI tool + MCP server. Describe what you want in plain English — FASTRACK parses your intent, builds a workflow from it, runs it against your connected tools, fixes itself when steps fail, and suggests improvements after a few successful runs.

• Model agnostic — Groq, OpenAI, Anthropic, Google, or any OpenAI-compatible endpoint

• Local memory — everything lives in SQLite at ~/.fastrack/fastrack.db. Nothing leaves your machine except the API calls you ask for

• Tool connectors — GitHub, Notion, Slack, Discord, Telegram, Linear, Airtable, Jira, Email (Resend), generic Webhooks

• Status reports — fastrack report pulls activity from your tools and delivers a formatted report anywhere

• Meeting notes → workflows — fastrack note digests notes into decisions and action items, then builds workflows from them

• Schedules — "every morning at 9am" becomes a real cron job

• MCP server — all commands exposed as tools for Claude Code, Cursor, or any MCP client

Install

npm install -g fastrack

For development: clone this repo and run npm link.

Requires Node.js >= 22.

Related MCP server: n8n Autopilot MCP SaaS

Quick start

Fastest path — zero config. Grab a free key at console.groq.com, then:

export GROQ_API_KEY=gsk_...        # PowerShell: $env:GROQ_API_KEY="gsk_..."
fastrack "summarize my open PRs"   # works immediately

Or run the interactive setup:

fastrack init

The interactive setup walks you through:

1. Picking a model provider and pasting your API key

2. (Optionally) connecting GitHub, Notion, or Slack

Then just talk to it:

fastrack "summarize my open PRs and add the summary to Notion"
fastrack "every morning at 9am summarize open PRs and add to Notion"
fastrack "what are the recent commits in my repo?"

Commands

How model agnostic works

All model configs live in ~/.fastrack/fastrack.config.json — never in the repo, never hardcoded. You can register multiple models and FASTRACK routes each request:

• Groq is first-class: pick it in fastrack init, or just set GROQ_API_KEY — no setup at all (defaults to llama-3.3-70b-versatile, override with FASTRACK_MODEL)

• Short + simple tasks (summarize, format, draft) go to the fastest/cheapest configured model

• Complex tasks (build workflow, debug, analyze) go to the smartest configured model

• fastrack model compare "..." runs the same prompt on all of them side by side

Custom provider? Any OpenAI-compatible endpoint works:

fastrack model add   # pick "custom", paste base URL + key + model name

No model account at all? Use the managed relay (deployed on Vercel by the project owner). It speaks the OpenAI protocol, so it plugs in as a custom provider:

fastrack model add
# provider: custom
# base URL: https://<relay-host>/v1
# API key:  <relay client token issued by the owner>

The relay holds the real Groq key server-side, so users never need their own provider account. See relay/README.md for deployment.

How memory works

• workflows — every workflow you create, its steps, trigger, run count and success rate

• context — facts FASTRACK picks up along the way

• history — every run: input, output, model used, duration, errors

The database lives at ~/.fastrack/fastrack.db. fastrack memory clear wipes it. That's the whole surface.

How workflow self-healing works

When a step fails, FASTRACK:

1. Sends the failed step, the error, and the workflow context to the model

2. Gets back corrected parameters (or corrected steps)

3. Retries — up to 3 attempts

4. If all attempts fail, the workflow is marked failed and you see exactly which step broke

After every 3rd successful run, FASTRACK analyzes the last 10 runs and offers an optimization. On the CLI it asks Apply? (y/n); over MCP it includes the suggestion in the response instead of blocking on a prompt.

Schedules

Recurring workflows need a long-running process — cron timers die when a process exits:

fastrack daemon     # dedicated scheduler process
fastrack mcp start  # the MCP server also runs schedules in the background

A one-shot CLI call parses and runs, but it cannot keep schedules alive by itself.

How to share workflows

fastrack workflows share 3        # writes fastrack-workflow-3.json to the current directory
fastrack workflows import fastrack-workflow-3.json

Shared files contain the workflow definition only — never your credentials.

MCP setup

Add to your Claude Code / Cursor MCP config:

{
  "mcpServers": {
    "fastrack": {
      "command": "fastrack",
      "args": ["mcp", "start"]
    }
  }
}

You then get these tools: fastrack_run, fastrack_workflow_list, fastrack_workflow_run, fastrack_workflow_share, fastrack_memory_show, fastrack_model_compare, fastrack_connect, fastrack_report, fastrack_note.

Configuration

Everything lives in ~/.fastrack/fastrack.config.json:

{
  "models": [{ "provider": "openai", "apiKey": "...", "model": "gpt-4o-mini" }],
  "activeModel": "openai",
  "connectors": { "github": { "token": "...", "owner": "...", "repo": "..." } },
  "preferences": { "autoSelect": true, "simpleModelThreshold": 500 }
}

This file holds API keys — it lives in your home directory, outside any git repo. fastrack.config.json is never created inside the project.

Testing

npm test

The suite runs fully offline: a fake OpenAI-compatible model server (test/helpers/fake-model.js) stands in for the LLM, and FASTRACK_HOME redirects all state (config + SQLite) to a temp directory, so your real ~/.fastrack is never touched.

License

MIT