Agentic Workflows MCP server
Commit Sep 21, 2026Guides AI coding agents through structured engineering workflows with plans, notes, and project memory.
Also known as codemcp/workflows
- Stars
- 79
- Tools
- —
- Downloads, 30 days
- —
- Latest version
- —
- Last commit
- Sep 21, 2026
- License
- GPL-3.0
What is Agentic Workflows MCP?
Agentic Workflows guides AI coding agents through structured engineering workflows with plans, notes, and project memory.
Summary generated from the project's README
About Agentic Workflows MCP server
Guiding a coding agent through a structured engineering process, with plans and notes that persist, is what this package does for developers who want more method than a quick prompt-and-patch loop. It ships a set of established workflows, including a classical V-model waterfall, Anthropic's EPCC, test-driven development and a reproduction-based bugfix process, and it keeps the agent aware of which phase it is in so it knows what step comes next. Notes and task plans written along the way survive context compression and even separate sessions, and project memory carries across conversations and branches, alongside automatic documentation and decision tracking.
Setup runs through the command line for agents such as kiro, claude, gemini, opencode and copilot, in a default config mode or a skill mode that loads on demand. It works with any client that speaks the Model-Context-Protocol, so switching between terminal UIs and IDEs does not change how you work.
The stated difference is emphasis on the engineer's own thinking: the tooling fills the conversation with material from your brain rather than assuming the agent inferred it. An experimental crowd feature, run with crowd-mcp, adds business-analyst, architect and developer roles with structured handoffs and messaging; commands such as agents copy and agents list set it up, with workflows like sdd-feature-crowd and sdd-greenfield-crowd.
What you can do with it
- 1Plan a greenfield card gameAsk the agent to start a UNO-like card game in an empty directory and have it plan tasks and take notes stage by stage instead of jumping straight to code.
- 2Keep notes across sessionsLet the agent record planning notes during a long task so they remain available after context compression or a new conversation.
- 3Follow a bugfix workflowHave the agent work a defect through the reproduction-based bugfix process with tracked decisions.
- 4Set up skill modeInstall the workflow guidance for a Claude agent using the on-demand skills format rather than the default config mode.
- 5Run specialized agent teamsCopy preconfigured analyst, architect and developer agents and let them hand off responsibilities on a feature.
Written from the project's README. Every use case is checked against a quote from it.
Automated checks
Ran Nov 14, 2026Full report →Not verified yet. A server needs a score of 80, no serious finding, and enough data for us to check. How checks work · What to fix for more points
- Source repository linked and reachablecodemcp/workflows is public and reachablegithub.com/codemcp/workflows ↗Pass
- Package names this repository as its sourceNot checked
- Build provenance (Sigstore attestation)Not checked
- Open source license (OSI)GPL-3.0spdx.org/licenses/GPL-3.0.html ↗Pass
- Commit or release in the last yearLast commit or release 2026-09-21github.com/codemcp/workflows ↗Pass
- OpenSSF ScorecardNot checked
- Package name not close to a popular packageNot checked
- No critical advisory in the package or its dependenciesNot checked
- No hidden instructions to the model0 tool descriptions and the README read; no override phrases, credential paths or hidden charactersPass
- Destructive tools are labelledNot checked
- Remote endpoint answers the MCP handshakeNot checked
README
View on GitHubAgentic Workflows
fka. "Responsible Vibe MCP"
Vibe coding is dead. Agentic Engineering is on the rise. Workflows have been here forever.
Transform any AI coding agent into a structured development partner with battle-tested engineering workflows.
⚡ Quick Start
# Setup your coding agent (config mode is the default)
npx @codemcp/workflows setup kiro # or claude, gemini, opencode, copilot
# Or use skills (agentskills.io format) for on-demand loading
npx @codemcp/workflows setup claude --mode skill # or gemini, opencode, copilot, kiro
Head over to a new empty dir and ask your agent: "Build a UNO-like card game" – and instantly experience how your agent doesn't just shoot, but starts engineering – with YOU in the driver seat!
🎬 See It In Action
What You Get
✅ Multiple battle-tested workflows (classical V-model "waterfall", Anthropic's EPCC, Test-Driven-Development, Reproduction-based bugfix, and many more)
✅ Context-aware process guidance: Your agent will take notes and plan tasks which survive context compression and even sessions.
✅ Project memory across conversations and branches
✅ Automatic documentation and decision tracking
✅ Multi-agent collaboration with specialized roles (business-analyst, architect, developer)
Universal MCP Support
Works in any agent that supports the Model-Context-Protocol. Whenever a new IDE or Terminal UI rises: You don't need to change the way you work.
How It's Different
There are zillions of "game changer next gen IDEs" out there, each claiming they will revolutionalize how software is going to be created.
In the end, they all just manipulate context of a transformer based LLM – it's all just noodle soup. And they are fast at changing what's going to be part of the context, and it's even getting accelerated by parallel agents.
The problem: The faster agents become, the harder it is to engineer. I strongly believe that software engineering is a creative process which majorly happens inside the engineers brain.
Responsible-Vibe-MCP helps to fill the conversation context with contents from YOUR brain – instead of relying the agent will have understood what you should have thought about.
And since may developers tend to through structured processing off the cliff once they think they start coding, the workflows server will maintain proactive process guidance - your AI knows what to do next in each development phase, follow proven engineering methodologies, and maintain long-term project context.
How It Works
Check the 📖 Complete Documentation →
There is also a recorded session on "how to tame your stubborn software agent" as part of the video podcast Software-Architektur.tv (German, auto-translated subtitles are okay-ish) which gives a more detailed insight into the basic ideas and how it's supposed to work.
⚠️ Experimental: Multi-Agent Collaboration (Crowd Workflows)
Enable teams of specialized AI agents to collaborate on development tasks with crowd-mcp:
- 👥 Team-based development: Business-analyst, architect, and developer agents work together
- 🔄 Structured handoffs: Clear responsibility transfers between agents
- 💬 Built-in collaboration: Agents consult each other via messaging
- 📋 Specialized workflows: sdd-feature-crowd, sdd-bugfix-crowd, sdd-greenfield-crowd
Quick Start:
# Copy pre-configured agent definitions
npx @codemcp/workflows@latest agents copy
# List available agent configurations
npx @codemcp/workflows@latest agents list
See Crowd MCP Integration Guide → for setup and usage.
Agentic Workflows MCP questions
How do I set up the server for my coding agent?
Run the setup command with your agent name, such as kiro, claude, gemini, opencode, or copilot, optionally using skill mode.
Does it keep project context across sessions?
Yes, it keeps project memory across conversations and branches and maintains notes and plans that survive context compression.