SkillFoxx

OysterWorkflow plugin

A Codex plugin and MCP bridge to an app that records your computer work and turns it into a graph of steps for an agent

  • 175 ★ on the repository
  • Last commit
  • Apache-2.0
Plugin

Broad access

Why this level

  • The app records the screen, visible text, clicks and keyboard input, and with voice comments the microphone too, so passwords and other people's data may end up in a recording
  • The agent performs scenario steps in real apps: it sends replies, edits data and works with mail and the browser
All reasons and checks

shuxinyang111/oysterworkflowAuthor:Shuxin Yang

Install

Manual install

codex plugin marketplace add ShuxinYang111/oysterworkflow

Then run codex plugin add oysterworkflow@oysterworkflow and start a new task. The OysterWorkflow app must be installed and running.

Access and checks

How we review

Broad accessThe tool can change data, run commands or handle secrets.

  • The app records the screen, visible text, clicks and keyboard input, and with voice comments the microphone too, so passwords and other people's data may end up in a recording
  • The agent performs scenario steps in real apps: it sends replies, edits data and works with mail and the browser

This is third-party code. Review the repository files before installing.

What it does

OysterWorkflow consists of a desktop app and an agent plugin. The app watches work on the computer: screen state, visible text, mouse clicks, keyboard input, app context and, optionally, voice comments. From the recording it extracts goals, branches, preferences, exceptions, recovery moves and completion conditions, and presents the result as a revisioned process graph that a person can review. The plugin connects that graph to an agent. For Codex it installs with two commands and starts a local STDIO bridge. The bridge finds the app's dynamic port through an owner-only connection file, so no port or token is configured by hand. For Claude, OpenClaw, Hermes and other MCP clients the README has a ready prompt: the agent itself adds the packaged `oysterworkflow-mcp` launcher as a local server. OysterWorkflow keeps the run state: graph, revision, transitions and retry limits. The agent performs the real actions in apps with the tools installed and authorized in it. The bridge tools are `search`, `fetch`, `prepare_workflow_run`, `get_workflow_run`, `advance_workflow_run` and `cancel_workflow_run`.

Who it is for. People who repeat the same multi-step computer processes, such as sorting inbound requests and drafting replies, and want to show them once and then hand them to Codex.

Good fit when

  • A routine passes through several apps and describing it in words takes longer than doing it once on a recording
  • The process has branches and exceptions that an ordinary macro does not survive
  • The recording should be reviewed as a graph of steps and corrected before an agent runs it
  • The agent already works in Codex on a Mac and needs a repeatable scenario with stops for approval
  • The same scenario has to be started from different MCP clients through a local bridge

Not a fit when

  • There is no Mac with Apple Silicon: the Codex plugin and the new step graph currently require one, and the Windows build is an earlier release
  • Commercial use: the desktop app code is under PolyForm Noncommercial, and business use needs a separate license
  • The screen may show other people's data, passwords or work secrets that must not be recorded
  • The scenario must run without the app: the plugin works only while OysterWorkflow is running

Example request

Use OysterWorkflow to run "Screen sales inquiries and prepare replies"

Limitations

  • The plugin and the latest step graph require macOS on Apple Silicon. The Windows x64 build is early, version 0.1.0.
  • Recording needs screen recording, accessibility and input monitoring permissions, and voice comments also need the microphone.
  • The app must stay running during execution, and every app in the scenario must be available to Codex itself.
  • The license is mixed: the core is Apache-2.0, the desktop product code is PolyForm Noncommercial 1.0.0. Check `LICENSING.md` before reusing files.

How to disable. In Codex turn off the oysterworkflow plugin in the plugin list and, if you like, remove the ShuxinYang111/oysterworkflow marketplace. For other clients delete the local `oysterworkflow` server from MCP settings. Stop recording and quit OysterWorkflow, and revoke the screen and input permissions in macOS system settings.

FAQ

How do I connect OysterWorkflow to Codex?

Install and start the app, then run `codex plugin marketplace add ShuxinYang111/oysterworkflow` and `codex plugin add oysterworkflow@oysterworkflow`. Start a new Codex task and ask it to run a scenario by name.

Does it work with Claude Code and other agents?

Yes, through the local MCP bridge. The README has a prompt the agent runs itself: it finds `oysterworkflow-mcp` in the app package and adds it as a stdio server. On macOS the path is usually `/Applications/OysterWorkflow.app/Contents/Resources/mcp/oysterworkflow-mcp`.

What does the app record?

Screen states, visible text, mouse and keyboard actions, app context and, optionally, voice. That needs screen recording, accessibility and input monitoring permissions. In support requests do not attach passwords, customer data or private addresses.

Can OysterWorkflow be used in commercial projects?

The Apache-2.0 core allows it, but the desktop app code is distributed under PolyForm Noncommercial 1.0.0. Commercial use of the app needs a separate license from the authors.

Which tools does the bridge expose?

Six: `search`, `fetch`, `prepare_workflow_run`, `get_workflow_run`, `advance_workflow_run` and `cancel_workflow_run`. The first two search and read scenarios, the rest prepare a run, show its state, advance it step by step and cancel it.

Discussion

No questions or reviews yet.

Plugins
Official

Zapier's official MCP plugin: the agent gets actions across thousands of apps through your Zapier account

PluginBroad access458

Open Dynamic Workflow turns repeatable agent tasks into validated scripts for Codex, Gemini, Cursor and other agent CLIs.

CLIExtended access84
Official

A self-hosted control plane for running several agents: task dispatch, run review, spend tracking, connected via MCP, CLI or REST

MCP serverBroad access6.5K
Official

A plugin from the n8n team: 13 skills and a bundled MCP connection teach Claude Code and Codex to build, edit and validate n8n workflows

SkillBroad access601

An MCP server for n8n: an assistant creates, runs and deletes workflows through the API, checks executions and tags and requests a security audit

MCP serverBroad access550

A Claude Code plugin and CLI for scientific facilities: an agent finds control channels, writes to hardware after approval.

PluginBroad access83

README in short

The README explains the idea of work experience for an agent and shows the path from recording to running in three steps. It then covers connecting to Codex and other MCP clients, downloading the app for macOS and Windows and the mixed licensing. The page ends with links to the roadmap and the rules for reports.

README badge

Report a problem