Back to Docs
New

Hosted execution

Run any MCPizy recipe end-to-end in your browser — no Claude Code install, no npx commands, no .claude.json. Our server spawns the MCPs, streams results live, and leans on our plan cache to keep token costs predictable.

No local setup

Click "Run" on any recipe. We spawn the MCPs on our infrastructure and stream events back to your browser.

Plan cache built in

Repeat runs reuse cached plans (Tier A). Arg-filling runs on Haiku while the heavy model stays idle — typically 60%+ token savings.

BYOK, session-only

Your Anthropic key and MCP credentials live in memory for the duration of the run, then disappear. Nothing is written to disk.

Which MCPs are supported

21 stateless, API-based MCPs are on the hosted whitelist. Stateful MCPs (filesystem, shell, docker) stay local by design.

🔮
Perplexity
Search
📝
Notion
Productivity
🧠
Anthropic
AI & ML
🤖
OpenAI
AI & ML
🔍
Tavily
Search
🔥
Firecrawl
Web Scraping
🦎
CoinGecko
Finance

How BYOK works

  1. 1You provide your own sk-ant-... Anthropic API key. We never proxy our own billing through your runs — you pay the underlying tokens directly to Anthropic.
  2. 2For each MCP the recipe uses, you provide that MCP's API credentials (e.g. a Notion integration token, a Stripe secret key). These stay in request memory only.
  3. 3The server spawns the MCP subprocesses with those env vars, drives the Anthropic agent loop, and streams the events back to your browser over SSE. When the stream closes, every credential is garbage-collected.

Plan quotas

PlanHosted executions / monthBest for
Free5Kick the tires, ship a demo
Pro100Individual builders, daily automations
Team1000Shared workspaces, production workflows

Quotas reset automatically on the first of each month. You always pay the Anthropic token cost for each run — the quota governs only how many hosted runs we spawn on your behalf.

Privacy + security

  • Anthropic API key + MCP credentials never land in the database or logs — they only exist in the request's Node process.
  • Tool outputs are redacted before persistence. We strip Bearer tokens, sk-* / pplx-* patterns, and any header/body keys named api_key, token, secret, etc.
  • Every execution row, step, and plan-cache template is scoped to your user_id. Row-level security enforces it — another tenant cannot read your runs.
  • Each execution is capped at 5 minutes wall-clock and 10 agent iterations. Client disconnect also aborts the run server-side.

Hosted vs local CLI

Use hosted when…

  • You want a one-click demo without installing anything.
  • The recipe only uses API-based MCPs on our whitelist.
  • You need plan-cache savings without running the proxy yourself.

Use local CLI when…

  • The recipe touches your filesystem, shell, or a local database.
  • You want Claude Code or Cursor as the driver instead of our UI.
  • Your compliance requires credentials never leave your machine.

Ready to try hosted execution?

Filter the recipe library to hostable-only and click Run.

Try a hosted recipe
💳
Stripe
Payments
💬
Slack
Communication
🐙
GitHub
Version Control
🦊
GitLab
Version Control
🔲
Linear
Productivity
📧
Resend
Communication
📬
SendGrid
Communication
🎙️
ElevenLabs
AI & ML
🛍️
Shopify
E-commerce
🐛
Sentry
Monitoring
🦔
PostHog
Analytics
⚡
Supabase
Databases
📚
Context7
Developer Tools
📚
DeepWiki by Devin
AI & ML