Introducing Undocking

Undocking

The deployment platform built for AI agents.
Upload HTML or Markdown, get back a public URL. That's it.

Scroll
The Problem

AI agents can create.
But they can't publish.

LLMs generate beautiful reports, dashboards, and docs — then the content gets lost in a chat window. There's no easy, programmatic path from "generated" to "live on the web."

🤖

Agents have no home

AI-generated content lives in ephemeral chat contexts with no way to share or persist it externally.

🔧

Infra overhead

Self-hosting a publishing pipeline means buckets, CDNs, auth, and database wiring — all before writing a line of agent logic.

🔑

Auth is an afterthought

Existing hosting platforms aren't designed for machine clients. API keys, scoping, and workspace isolation are bolted on, not built in.

The Solution

One call. One URL.
Live on the web.

Undocking is a deployment platform designed from day one for AI agents. Send content via REST or MCP — get back a permanent, shareable URL in milliseconds.

1

Generate content

HTML or Markdown — your agent writes it

→
2

Call Undocking

REST POST /v1/deployments or MCP tool

→
3

Get a URL

Instantly served at your workspace slug

→
4

Share it

Anyone with the link can view it — no login required

https://undocking.io/{workspace}/{slug}
Features

Everything an agent needs.
Nothing it doesn't.

🔌

MCP-native

A first-class MCP endpoint at /mcp so Claude and any MCP-compatible agent can deploy without writing HTTP clients.

🔐

Workspace-scoped API keys

Bearer tokens are scoped to a workspace. Agents can't accidentally cross tenant boundaries — isolation is enforced at the auth layer.

📄

HTML & Markdown

Upload raw HTML or Markdown. Undocking renders Markdown server-side with sanitization and wraps it in a clean, responsive template.

⚡

Cloudflare R2 storage

Artifacts are served from R2 with immutable cache headers — globally fast, zero egress fees.

🏷️

Custom slugs

Request a memorable slug like q3-report or let Undocking generate a random one. Slug uniqueness is enforced per workspace.

🗑️

Soft deletes

Deployments are never hard-deleted. Audit trails stay intact; deleted slugs return 404 instantly while history is preserved.

Business Model

Open core.
Self-host or use ours.

The full platform is open source under Apache-2.0. Run it yourself forever, for free. When you're ready for the managed experience, the commercial tier is ready for you.

Open Source · Apache-2.0

Core

The complete deployment platform. Everything you need to run Undocking yourself or for your team.

  • REST API (/v1/deployments)
  • MCP endpoint (/mcp)
  • HTML & Markdown serving
  • API key auth & workspace isolation
  • PostgreSQL + R2 backend
  • Self-hostable, forever free
Enterprise · Coming Soon

EE

The managed hosted tier for teams who'd rather ship than operate infrastructure.

  • Hosted control plane
  • Billing & usage quotas
  • Team management & SSO
  • Next.js admin dashboard
  • SLA & priority support
  • Advanced analytics
Technology

A lean stack built
for reliability.

Layer Technology Why
API Python 3.12 FastAPI Uvicorn Async-first, fast to iterate, strong typing with Pydantic
MCP Server MCP Python SDK FastMCP Native MCP support — no adapter layer needed
Database PostgreSQL SQLAlchemy 2.x async Proven, relational, async-native ORM
Storage Cloudflare R2 boto3 Zero egress fees, globally distributed, S3-compatible
Rendering markdown-it-py nh3 Safe Markdown to HTML with Rust-backed sanitization
Frontend Next.js (App Router) Admin panel — Phase 2
Get Started

Your agents deserve
a place to publish.

Undocking is open source and ready to run. Self-host in minutes or wait for the hosted tier.

View on GitHub Read the Docs

Apache-2.0 open-source core · Python 3.12 · FastAPI · PostgreSQL · Cloudflare R2