Lewati ke konten
TakoAPI
Kembali ke skill

n2-stitch-mcp

oleh choihyunsus

AI & LLMs
559Unduhan dilaporkan oleh clawskills.sh
0Tayangan
0 sukaMasuk untuk menyukai skill ini
—Belum ada penilaian
Penilaian Anda

Ringkasan

Resilient MCP proxy for Google Stitch — 3-layer safety (auto-retry, token refresh, TCP drop recovery).

Deskripsi

Apa yang Dilakukan Skill Ini

Google Stitch's screen generation calls take 2-10 minutes, but the API drops TCP connections after ~60 seconds, causing most MCP proxies to silently lose the result. N2 Stitch MCP recovers by switching to auto-polling after a drop, tracks each generation's state, and delivers results when complete. Token refresh runs in the background and failed requests retry with exponential backoff.

Unlike other Stitch MCP servers that fail silently when TCP drops during generation, this proxy polls for completion and returns the result regardless of connection interruptions.

Kapan Menggunakannya

  • Generating UI screens from text without losing long-running Stitch jobs
  • Checking screen generation progress during a 2-10 minute Stitch call
  • Managing Stitch projects and screens through Claude
  • Editing existing Stitch screens and generating design variants
  • Running Stitch API calls that would normally fail at the 60-second TCP limit

Contoh Alur Kerja

INPUT

User asks: Generate a dashboard screen with a sidebar navigation and data charts

AGENT
  1. 1

    Calls generate_screen_from_text with the description

  2. 2

    TCP connection drops after ~60 seconds as expected

  3. 3

    Switches to polling via generation_status tool

  4. 4

    Detects generation completion and fetches the result

  5. 5

    Returns the generated HTML/CSS to the user

OUTPUT

Full HTML/CSS for the dashboard screen delivered despite the TCP drop

Persyaratan

  • Google Cloud account with Stitch API access (via gcloud auth application-default login), or a Stitch API key set as the STITCH_API_KEY environment variable

Instalasi

clawhub install choihyunsus/n2-stitch-mcp

Jalankan perintah CLI-nya, atau tempel prompt ke Claude Code, Cursor, Codex, atau coding agent apa pun yang Anda pakai.

Tautan