CC Workflow Studio
Design AI agent workflows on a visual canvas and export them as Markdown skills and slash commands your coding agent already understands.
- Source repo
- breaking-brake/cc-wf-studio
- Stars
- ★ 5.4k
- Last updated
- 14d ago
- License
- NOASSERTION
- Primary language
- TypeScript
- FA score
- 50/100 · Major gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformChatGPT · Codex · Claude Code
- Cost
- Free software; you pay for model usage
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- A Claude Code user who wants to turn a multi-step prompt chain into a reusable slash command or sub-agent under .claude/.
- Not a fit if
- Teams that don't use any AI coding IDE or CLI
- Users who want a web-only canvas without installing VS Code
- Teams that require a closed SaaS-only workflow platform
- Source review
- 50/100 · Major gaps
What does this agent do, and when should you use it?
CC Workflow Studio turns AI workflow authoring from prompt guesswork into a visual design task: you drag nodes onto a React Flow canvas, wire up sub-agents, Agent Skills and MCP tools, then export the result into the file layout that Claude Code, GitHub Copilot Chat, Copilot CLI, OpenAI Codex CLI, Zoo Code, Gemini CLI, Antigravity or Cursor already reads. The primary entry point is the VS Code extension breaking-brake.cc-wf-studio, which stores workflows as workflow.json under .vscode/workflows/. That same file can be read and written by @cc-wf-studio/cli (ccwf render | validate | export | run | preview | canvas | mcp) and by the @cc-wf-studio/mcp server (stdio command ccwf-mcp), so terminals, CI, SSH, Codespaces and external MCP clients can all take part. An Edit with AI panel launches the MCP server behind the scenes, letting an AI agent call get_workflow and apply_workflow to change the canvas live. The extension is AGPL-3.0-or-later while the reusable core, mcp and cli npm packages are MIT.
Reads and writes workflow files kept as .vscode/workflows/*.json; lets you add and configure canvas nodes, then save and load them through toolbar buttons; exports to per-agent destinations including .claude/agents/, .claude/commands/, .github/prompts/, .github/skills/, .codex/skills/, .roo/skills/, .gemini/skills/, .agent/skills/, .cursor/agents/ and .cursor/skills/ as Markdown agent skills or slash commands; runs a workflow directly inside a coding agent via the Run button; in Edit with AI mode starts @cc-wf-studio/mcp so an agent can call get_workflow to fetch the JSON and apply_workflow to push edits back to the canvas; and through the ccwf CLI subcommands render, validate, export, run, preview, canvas and mcp, performs the same operations without an open VS Code window.
- A Claude Code user who wants to turn a multi-step prompt chain into a reusable slash command or sub-agent under .claude/.
- A team whose members split across Gemini CLI, Codex CLI and Copilot and need one workflow exported into each tool's own skill directory.
- A developer working over SSH, in Codespaces or in CI who cannot open a VS Code canvas and instead runs npx @cc-wf-studio/cli to validate and render.
- Someone who prefers asking Claude Code or another MCP client to generate and refine a workflow in natural language rather than dragging nodes.
- A reviewer who needs to audit sub-agent responsibilities and MCP tool ordering by looking at the visual graph instead of raw prompts.
- A team moving away from Dify-style orchestration toward file-based Markdown workflows kept in the repository.
How do you install or deploy this agent?
Install the extension from the Marketplace on any machine with VS Code:
code --install-extension breaking-brake.cc-wf-studioIt is also published on the VS Code Marketplace and Open VSX under cc-wf-studio. To work only from a terminal, no extension is needed:
npx @cc-wf-studio/cli --helpThe MCP server ships as the npm package @cc-wf-studio/mcp with the stdio name ccwf-mcp; it needs a Node.js runtime and must be registered in your MCP client manually (the README only says to add it to .mcp.json and does not publish a ready-to-copy snippet, so consult the packages/mcp docs for the exact keys). Whichever export targets you use — Claude Code, Copilot, Codex, Zoo Code, Gemini CLI, Antigravity or Cursor — that tool must be installed separately.
How do you use this agent?
After installing the extension, click the CC Workflow Studio icon in the top-right corner of the editor or open it from the Command Palette:
CC Workflow Studio: Open EditorAdd nodes from the palette, configure them, then press Save to store the workflow at .vscode/workflows/*.json. Press Export to generate a Markdown slash command or agent skill for the selected target (agents other than Claude Code are activated from the toolbar's More menu), then invoke it inside your coding agent as /workflow-name.
To operate on the same file from a terminal:
npx @cc-wf-studio/cli validate workflow.json
npx @cc-wf-studio/cli preview workflow.json
npx @cc-wf-studio/cli export workflow.jsonFor Edit with AI, click an agent button in the panel; the extension starts the MCP server automatically and the agent iterates through get_workflow and apply_workflow against the live canvas.
What are this agent's strengths and limitations?
- One workflow.json is shared by the VS Code extension, the CLI and the MCP server, so the canvas, the terminal and external AI clients operate on the same artifact.
- Export targets eight named agents — Claude Code, Copilot Chat, Copilot CLI, Codex CLI, Zoo Code, Gemini CLI, Antigravity and Cursor — each using that tool's own native Markdown directory convention.
- Edit with AI is a real round trip through MCP get_workflow / apply_workflow, so AI edits land back on the canvas instead of producing throwaway text.
- Licensing is split deliberately: the extension is AGPL-3.0-or-later while core, mcp and cli on npm are MIT, so libraries can be reused in other projects.
- The CLI exposes render, validate and preview subcommands, making workflows checkable in CI or Codespaces without a GUI.
- Every target except Claude Code must be activated manually from the toolbar's More menu rather than through a default one-click export.
- Registering the MCP server in a client requires you to write the .mcp.json entry yourself; the README gives no complete copyable configuration.
- The extension itself is AGPL-3.0-or-later, which constrains closed-source embedding; teams must check whether the MIT npm packages alone are enough.
- Workflows are only executed by an external coding agent, so results depend on that agent and model, and you must supply the corresponding subscription or API key.
- The Usage Examples section is marked Coming soon, so there are no ready-made sample workflows or tutorials to learn from.
How does this agent compare with similar options?
The README credits Dify as inspiration; Dify is a visual LLM workflow orchestration platform, whereas CC Workflow Studio's differentiator is that output lands as Markdown skills and slash commands inside your repository directories and stays consumable by its CLI and MCP server.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| CC Workflow Studio This agent | 50 · Major gaps | IDE extensionFree + model costs | ★ 5.4k | 14d ago | TypeScript | ChatGPT · Codex · Claude Code |
| MobileBuildMCP | 49 · Major gaps | CLIFree | ★ 6.5k | 3d ago | TypeScript | Codex · Claude Code |
| Microsoft Learn MCP Server | 43 · Major gaps | MCP serverFree | ★ 1.9k | 1mo ago | TypeScript | ChatGPT · Codex · Claude Code · Claude.ai |
| agnix Config Linter | 89 · Good | CLIFree | ★ 455 | 2d ago | Rust | Codex · Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
SECURITY.md asserts sandboxed webview, CSP, input validation, local-first design, and conflict detection with user confirmation before overwriting, supporting user_confirmation at 2. However least_privilege is only asserted, with no concrete permission manifest or minimal-privilege implementation; data_flow_transparency only vaguely notes MCP network tools need review without data-flow detail; sensitive_data_handling has no key/secret handling guidance; external_effects covers exporting to .claude/ etc. and running workflows but lacks rollback or blast-radius description; rollback only mentions overwrite confirmation, no version revert; source_attribution only thanks React Flow/Dify without third-party code provenance. dependency_security has Snyk, Dependabot, frozen-lockfile and CI gates, so 2.
self_consistency: README, package.json and CI agree on monorepo structure and build order, 2. dependency_availability: pnpm lock, Node>=20, CI install steps, 2. failure_messages: no visible error-message design in source, only conflict detection mentioned in SECURITY.md, so 1.
audience_and_scenarios clearly lists multiple AI agents and three entry points (CLI/MCP/extension), 2. capability_boundaries defined via Supported Agents table and Requires column, 2. trigger_precision only mentions command palette and toolbar buttons without precise trigger conditions, 1. environment_fit covers VSCode, CLI, SSH, Codespaces, 2.
information_architecture has clear sections and package structure, 2. install_notes gives extension install, npx CLI, MCP config, 2. naming_stability has no versioned naming or stability commitment, 1. examples_and_faq explicitly states Usage Examples are Coming soon, 1. known_limitations documented in SECURITY.md for network and file system access, 2. license has full AGPL-3.0 text plus per-package MIT note, 2. versioning_changelog has changesets, release workflow and CHANGELOG mention, 2. maintenance_responsibility: publisher unverified, responsibility only via SECURITY.md response-time promise, 1.
output_usability exports .md commands/skills that run directly, 2. marginal_value offers visual orchestration and multi-agent export over hand-written prompts, 2. cost_benefit requires extension install, MCP config and multiple external agents with unquantified benefit, 1.
claim_traceability: most claims (sandbox, CSP, Snyk scans) lack corresponding source or config evidence, only CI and package.json partially corroborate, 1. cross_source_corroboration: README, SECURITY, CI partially agree but security claims lack cross-evidence, 1. fact_inference_separation: docs mix assertions with facts without distinguishing verified vs inferred, 1.
- Publisher identity is unverified; maintenance responsibility and update path are unclear, so verify provenance before adoption.
- Security claims (sandbox, CSP, Snyk scans) are largely documentation assertions without corresponding source or config evidence; independent audit needed.
- Exporting and running workflows writes to .claude/ etc. and may trigger external agent execution, with no rollback mechanism described.
- Usage Examples are marked Coming soon, lacking reproducible examples and FAQ.
- License is AGPL-3.0 with per-package MIT; commercial use and redistribution require attention to the compliance difference.