An Agent Skill that lets coding agents write and run custom Playwright browser automation scripts for tests and multi-step web tasks.
reticlehq/reticle
A dev-only SDK and MCP server that lets AI coding agents drive your running web or desktop app and return a pass, fail, or cannot-tell verdict with evidence.
- Stars
- 1,276
- Forks
- 220
- Open issues
- 220
- Last commit
- 16 hours ago
What is it?
Reticle addresses a common gap in AI-assisted coding: an agent writes a change, assumes it works, and never opens the app. Reticle installs a small development-only SDK inside your app and exposes tools over MCP so the agent can look at, act on, observe and assert against the running application.
Because the SDK reads from inside the page, it can see network responses tied to the click that caused them, store state from libraries such as zustand, Redux, TanStack Query and Pinia, console errors, and the source file and line of each element. Every check ends in a verdict of yes, no, or unknown, with the evidence attached. Saved flows can be replayed later without an AI model, and `reticle gate` can be used in CI to require a passing run when covered code changes.
Reticle supports React, Next.js, Vue, Svelte, Angular and similar web stacks, plus Electron and Tauri desktop apps, and works with Claude Code, Cursor, Codex, Windsurf, VS Code, Gemini CLI and other MCP clients. The bridge listens on localhost only, and production builds exclude the SDK.
Key features
- 01Gives coding agents tools to look at, act on, observe and assert against a running app.
- 02Returns yes, no, or unknown verdicts with the request, state change and file:line as evidence.
- 03Reads network responses, store state, console output and React renders from inside the page.
- 04Replays saved flows deterministically without calling an AI model.
- 05Provides `reticle gate` for CI to require passing runs when covered code changes.
- 06Excludes the SDK from production builds and binds the local bridge to 127.0.0.1.
Use cases
- Checking that an agent-made change to checkout or login actually works in the running app.
- Catching silent failures such as a 500 response behind a success message.
- Locking critical user journeys so later edits cannot quietly break them.
- Requiring verification of covered flows in CI before merging.
Quick start
- 1
Install the reticle command globally with npm
npm install -g @reticlehq/server - 2
Register Reticle with your coding agents
reticle setup mcp - 3
Restart your coding agent, then run init in your app's folder
reticle init
Things to know
- The server and CLI are source-available under FSL-1.1-ALv2, not OSI-approved open source, and convert to Apache-2.0 after two years.
- It needs Node 20.11 or newer, a locally running dev server, and a coding agent that supports MCP.
- It drives a Chromium browser, and anonymous usage telemetry is on by default unless you turn it off.
Topics
FAQ
- What is reticle?
- A dev-only SDK and MCP server that lets AI coding agents drive your running web or desktop app and return a pass, fail, or cannot-tell verdict with evidence. Reticle addresses a common gap in AI-assisted coding: an agent writes a change, assumes it works, and never opens the app. Reticle installs a small development-only SDK inside your app and exposes tools over MCP so the agent can look at, act on, observe and assert against the running application.
- Is reticle free to use? What license does it have?
- reticle is open source under the Other license. Check the license terms before using it in commercial products.
- Is reticle still maintained?
- Actively maintained. The last commit to reticle was on Oct 21, 2026.
- What can I use reticle for?
- Checking that an agent-made change to checkout or login actually works in the running app. Catching silent failures such as a 500 response behind a success message. Locking critical user journeys so later edits cannot quietly break them. Requiring verification of covered flows in CI before merging.
- How do I install reticle?
- Install the reticle command globally with npm: npm install -g @reticlehq/server Register Reticle with your coding agents: reticle setup mcp Restart your coding agent, then run init in your app's folder: reticle init
- Where can I get reticle?
- The source code and installation instructions are on GitHub: https://github.com/reticlehq/reticle
Summary generated by AI from the project's README. GitHub stats refresh daily.
Related projects
A plugin-based harness that gives Claude Code, Codex and other coding agents shared agents, skills, hooks, memory and security scanning.
An Electron-based desktop browser for testing websites across many device previews at once, with mirrored interactions and an MCP server for AI agents.
A local catalog and control plane that lets Codex or Claude search 2,700+ SKILL.md playbooks, pick exact skills, validate a stack and preview an install plan.
Plain-language agent skills for Claude Code, Codex CLI, Cursor CLI and Gemini CLI that turn session reports into a one-screen verdict anyone can follow.
A Claude Code skill that turns a website URL into a 60 fps product launch video built from HTML, with music-synced sound effects and motion blur.