Loading
Pagenta is a remote MCP server. Claude, ChatGPT, Cursor, and terminal agents can publish, lock, schedule, update, and measure real apps through it — in one round trip. No repo. No build pipeline. No human handoff.
Before
Agent writes HTML. You copy-paste it into Vercel, set up a repo, configure env vars, fight with build errors.
With Pagenta
Agent writes HTML. One MCP call. Live URL in 2 seconds.
Before
Need to update? Re-prompt the agent, copy the new HTML, redeploy, watch CI.
With Pagenta
Agent edits the live page in place. Same URL. No build step.
Before
Want auth, scheduled refresh, push notifications, analytics? Wire each one — Auth0, Vercel Cron, OneSignal, Plausible.
With Pagenta
All bundled. The agent calls Pagenta's MCP tools for each. No SaaS sprawl.
Before
Pray the agent doesn't ship something private to the open internet.
With Pagenta
Approval gates for sensitive actions. Audit log for everything else.
These aren't prompts to your tool — they're tools the agent calls. The human stays in the conversation; the agent calls Pagenta's MCP and the URL is live.
Replaces: copy-pasting HTML into Vercel
Tell Claude or Cursor to ship the HTML it just wrote — Pagenta's MCP server takes the file, gives it a URL, password, and analytics in one round trip. The agent never leaves your chat window.
Replaces: GitHub → CI → Vercel
Ask the agent to fix a typo or add a chart. It edits the live page in place at the same URL — no commit, no PR, no waiting for a build. The cron, secrets, and access rules persist.
Replaces: a cron job you set up yourself
"Pull yesterday's Stripe revenue and update this dashboard every morning." The agent writes the refresh script, stores the API key as a secret, schedules it on Pagenta. You see fresh numbers without writing a Lambda.
Replaces: blind trust in an LLM with API keys
Give your agent a $50/mo budget, scoped to one Space. Public publishes require your approval. Audit log shows every action. You ship fast — the leash stays short.
Add the MCP URL to Claude / ChatGPT / Cursor / Codex. See /mcp for the 30-second per-client setup.
Ask in plain English: "deploy this, password it, schedule a refresh, tell me when someone opens it." The agent calls Pagenta tools, you get a URL back.
Agent observes analytics, revises the page in place, updates the schedule. You stay in the conversation; the work goes live.
$12/mo (or $120/yr). 50 live apps, 5-minute cron, 10 secrets, 90-day analytics. Agent gets its own credits. Upgrade to Studio when you add custom domains, white-labeling, or more Spaces. Agency adds agent budgets + approval workflows for teams.
Anything that speaks MCP: Claude.ai (Connectors), Claude Desktop, Claude Code, Cursor, Codex, Goose, Windsurf, Gemini CLI, plus any future MCP-compatible agent. Also a direct CLI for terminal agents and plain scripts.
Deploy pages, attach them to Spaces, set passwords and expiry, schedule cron jobs, store secrets, send push broadcasts, read analytics, run experiments, and revise live pages — all via MCP tool calls. See /mcp for the full surface.
On Studio and above, agents can have their own subscription with capped credits. Sensitive actions (public publishes, secret changes, deletes, broadcasts) require explicit human confirmation. Audit log captures every action with its initiator.
Yes — get_page_analytics returns views, visitors, session length, scroll, CTA clicks, form submits, and per-variant experiment performance. An agent can loop on "deploy → observe → revise" without you in the loop.
Yes. The free Pagenta tier has 5 live apps, daily cron, basic analytics — plenty to wire an agent up and prove the loop. Upgrade to Pro when the agent ships often enough to bump against the cap.
pagenta.ai/api/mcp — same server for all clients. Add it to your agent's config (see /mcp for per-client instructions) and Pagenta's tool surface shows up in the agent's tool list.
Free tier is real. Wire your agent in one round trip and watch it ship the first live URL inside your existing chat.