agentclip 0.3.0

Turn AI agent QA runs into shareable slideshows.

agentclip

Turn your AI agent's QA runs into shareable clips.

Your AI agent drives the browser. AgentClip captures the moments that matter, screenshots and captions, and ships you back a single URL anyone can watch.

The product is the artifact (the clip URL) plus the prompt engineering (the bundled skill) that makes generated runs actually good. The backend, MCP plumbing, and CLI are commodity.

Live demo: https://agentclip.dev

Status

Built in May 2026. Early. APIs may shift before a 1.0 tag.

Install

pip install agentclip

That's it. The first time agentclip runs, lazy first-run setup wires the bundled Claude Code skill, downloads browser drivers if you'll need them, and writes a marker so subsequent invocations are instant.

Or run with no install:

uvx agentclip --help

60-second example

Ask your agent in plain English (Claude Code, Claude Desktop, or any MCP-aware client):

QA the signup flow on localhost:3000 and post a clip.

The agent calls slideshow_create, captures screenshots after each meaningful action, calls add_slide with active-voice captions, then set_summary at the end. You get back a share URL plus an edit URL for caption fixes later.

Or use the CLI directly:

agentclip slideshow create --title "Signup QA" --description "fresh-user flow"
agentclip slideshow add <slideshow_id> /tmp/shot1.png --caption "Clicked Sign Up. Form posted."
agentclip slideshow summary <slideshow_id> "Signup passed. One real bug at slide 4."

# Manage what you've made:
agentclip slideshow list
agentclip slideshow delete <slideshow_id>

MCP install (Claude Desktop, Claude Code, Cursor)

Add to your claude_desktop_config.json (or equivalent):

{
  "mcpServers": {
    "agentclip": {
      "command": "uvx",
      "args": ["agentclip-mcp"]
    }
  }
}

Restart your agent runtime. The four tools below will register automatically.

Tools

Tool	Description
slideshow_create	Start a clip. Returns id, share URL, and a write_token used for subsequent mutations.
slideshow_add_slide	Append a screenshot + caption. The local state store auto-supplies the write_token.
slideshow_update_slide	Replace the image and/or caption of a slide already in the clip.
slideshow_set_summary	Set the clip's TL;DR, near the end of the run.

Configuration

Env var	Purpose	Default
AGENTCLIP_BASE_URL	Backend URL. Override for self-hosted.	https://agentclip.dev
AGENTCLIP_STATE_PATH	Write-token cache file.	~/.agentclip/state.json

Layout

• src/agentclip/sdk.py: AgentClipClient, sync HTTP client over httpx
• src/agentclip/cli.py: agentclip ... Typer CLI, thin wrapper over the SDK
• src/agentclip/mcp_server.py: MCP server registering the four tools
• src/agentclip/state.py: atomic-write ~/.agentclip/state.json for write_tokens
• src/agentclip/skill/SKILL.md: the bundled agent skill

Self-hosting the backend

Companion repo: ericelizes1/agentclip — the platform monorepo with the Django API and Next.js frontend. Reference deploy is Fly.io apps + Neon Postgres + Cloudflare R2 object storage, with secrets sourced from 1Password; docker compose up --build covers any Docker host. Point AGENTCLIP_BASE_URL at your domain. Full guide at docs.agentclip.dev/self-hosting.

Development

uv sync --all-extras --dev
uv run pytest -q
uv run ruff check .

See CONTRIBUTING.md for the contributor guide. CI runs lint + format + tests across Python 3.11/3.12/3.13 on every PR.

License

MIT. See LICENSE.