Smart Agent Workflow MCP
# Smart Agent Workflow MCP
> The only MCP that **enforces tests before merge**. Full-cycle development automation with testing gates and knowledge graph memory.
[](https://www.npmjs.com/package/smart-agent-workflow-mcp)
[](https://opensource.org/licenses/MIT)
## Why This MCP?
5+ git worktree MCPs exist. None require tests. **This one does.**
| Feature | Others | Smart Agent Workflow |
|---------|--------|---------------------|
| Worktree management | ✅ | ✅ |
| **Mandatory E2E tests** | ❌ | ✅ |
| **Build verification** | ❌ | ✅ |
| **Auto-rollback** | ❌ | ✅ |
| **Memory persistence** | ❌ | ✅ 🧠 |
| Full-cycle workflow | ❌ | ✅ |
## Quick Start
```bash
npx smart-agent-workflow-mcp
```
## Installation
```bash
npm install -g smart-agent-workflow-mcp
```
## Configuration
### Claude Desktop
Add to your `claude_desktop_config.json`:
```json
{
"mcpServers": {
"smart-agent-workflow": {
"command": "npx",
"args": ["smart-agent-workflow-mcp"]
}
}
}
```
### Claude Code
Add to your MCP settings:
```json
{
"mcpServers": {
"smart-agent-workflow": {
"command": "npx",
"args": ["smart-agent-workflow-mcp"]
}
}
}
```
## Tools
### Worktree Management
| Tool | Description |
|------|-------------|
| `create_worktree` | Create isolated feature environment with ephemeral worktree |
| `worktree_status` | List all active worktrees with metadata |
| `cleanup_worktree` | Merge and cleanup (requires tests to pass!) |
| `abort_worktree` | Cancel without merge |
### Usage Examples
#### Create a new worktree
```
Use create_worktree with task="Add user authentication" and base_branch="main"
```
#### List active worktrees
```
Use worktree_status
```
#### Complete and merge a feature
```
Use cleanup_worktree with worktree_path="/path/to/worktree"
```
#### Cancel a feature
```
Use abort_worktree with worktree_path="/path/to/worktree" and reason="Requirements changed"
```
## Resources
| Resource | Description |
|----------|-------------|
| `worktree://status` | JSON with current status of all worktrees |
## Philosophy
> "No merge without green tests. Every feature builds collective knowledge."
**The 3 Pillars:**
1. 🧪 **Test-First**: Quality gates that can't be bypassed
2. 🧠 **Memory**: Context that grows with your project
3. 📝 **Auto-Docs**: Documentation enriched by history
## Roadmap
| Version | Features |
|---------|----------|
| ✅ v0.1.0 | Worktree management |
| ✅ v0.2.0 | Testing integration (Playwright, build verification) |
| ✅ v0.3.0 | Full workflow (start_feature, complete_feature, rollback) |
| ✅ v0.4.0 | Documentation (auto-update CLAUDE.md, reports) |
| ✅ v0.5.0 | Memory (knowledge graph, context persistence) |
| ✅ v0.6.0 | Context Health (statusline, auto-checkpoint) (current) |
| 🔜 v1.0.0 | Production ready |
## Context Health (v0.6.0)
Smart Agent Workflow now tracks context health to anticipate compaction:
### Get Health Status
```
Use get_context_health
```
Returns health score (0-100), status (good/warning/critical), and recommendations.
### Manual Checkpoint
```
Use checkpoint_context with message="Before major refactor"
```
Saves current workflow state and resets health to 100%.
### Statusline Integration
Add to your Claude Code settings:
```bash
claude config set -g statusLine '{"type":"command","command":"npx smart-agent-workflow-mcp --statusline"}'
```
Shows: `Phase: testing | Feature: auth | Progress: 4/8 | Health: 75%`
## Development
```bash
# Clone the repo
git clone https://github.com/vjrivmon/smart-agent-workflow-mcp.git
cd smart-agent-workflow-mcp
# Install dependencies
npm install
# Run in development mode
npm run dev
# Build
npm run build
# Test
npm test
```
## Contributing
Contributions are welcome! Please read our [Contributing Guide](CONTRIBUTING.md) first.
## License
MIT © [Vicente Rivas Monferrer](https://github.com/vjrivmon)
---
Made with ❤️ by [Vicente Rivas Monferrer](https://vicenterivasmonferrer.dev)
## Recent Changes
Test entry from validation
TDQS
Scored across 20 tools
Several tools overlap in scope: abort_worktree, rollback_feature, and cleanup_worktree all handle worktree termination at different levels, while create_worktree and start_feature both create worktrees. The descriptions help clarify the hierarchy, but an agent may struggle to choose between low-level and high-level equivalents.
Most tools follow a consistent verb_noun snake_case pattern (e.g., start_feature, verify_build, save_context). The exception is 'worktree_status', which is a noun phrase rather than a verb-led command, making it slightly inconsistent with the rest.
At 20 tools, the server is on the heavier side of the typical range. While each tool has a distinct role, some are redundant due to nested operations (e.g., abort_worktree vs. rollback_feature), suggesting the count could be trimmed without losing functionality.
The tool surface covers the full feature development lifecycle—start, test, build, merge, rollback, and cleanup—plus memory/context management and documentation. Minor gaps exist, such as no direct way to list completed features or delete specific memories, but these are easily worked around.