https://github.com/diegonogueiradev/mcp-graph-workflow
MCP local-first CLI tool that converts PRD text files into persistent execution graphs (SQLite) for structured agentic workflows
https://github.com/diegonogueiradev/mcp-graph-workflow
agent-tooling agentic-workflows ai-tools cli local-first mcp mcp-server model-context-protocol open-source prd rag semantic-search sqlite task-graph typescript
Last synced: 19 days ago
JSON representation
MCP local-first CLI tool that converts PRD text files into persistent execution graphs (SQLite) for structured agentic workflows
- Host: GitHub
- URL: https://github.com/diegonogueiradev/mcp-graph-workflow
- Owner: DiegoNogueiraDev
- License: mit
- Created: 2026-03-04T18:57:56.000Z (3 months ago)
- Default Branch: master
- Last Pushed: 2026-04-13T22:48:03.000Z (about 2 months ago)
- Last Synced: 2026-04-13T23:29:32.421Z (about 2 months ago)
- Topics: agent-tooling, agentic-workflows, ai-tools, cli, local-first, mcp, mcp-server, model-context-protocol, open-source, prd, rag, semantic-search, sqlite, task-graph, typescript
- Language: TypeScript
- Homepage: https://mcp-graph-workflow.vercel.app
- Size: 52.7 MB
- Stars: 28
- Watchers: 1
- Forks: 4
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- Changelog: CHANGELOG.md
- Contributing: CONTRIBUTING.md
- Funding: .github/FUNDING.yml
- License: LICENSE
- Code of conduct: CODE_OF_CONDUCT.md
- Codeowners: CODEOWNERS
- Security: SECURITY.md
- Support: SUPPORT.md
Awesome Lists containing this project
README
mcp-graph
---
**mcp-graph** turns your PRD into a persistent task graph that your AI agent navigates instead of improvising.
PRD → graph (SQLite) → TDD → production. 100% local, zero cloud.
---
## The problem
Every AI coding session has the same three problems:
- **The agent forgets** — each new chat starts from zero, the plan gets reinvented
- **PRDs become walls of text** — nobody re-reads them, the agent improvises
- **No traceability** — you can't tell what was done, what's blocked, or why
## How it solves them
1. `import_prd` — reads the PRD and creates tasks with acceptance criteria in the graph
2. `start_task` — agent picks the next task with compressed context in hand
3. `finish_task` — validates DoD (12 checks), status → done, suggests the next
Everything lives in `workflow-graph/graph.db` — local SQLite, gitignored, never in the cloud.
## Installation
```bash
npm install -g @mcp-graph-workflow/mcp-graph
```
Add to your project's `.mcp.json` (Claude Code, Cursor, Copilot):
```json
{
"mcpServers": {
"mcp-graph": {
"command": "npx",
"args": ["-y", "@mcp-graph-workflow/mcp-graph"]
}
}
}
```
Initialize in your project:
```bash
mcp-graph init
```
**Prerequisites:** Node.js ≥ 18. No Docker. No LLM API key required.
## Getting started
```bash
# 1. Import a PRD (or create tasks manually)
mcp-graph import my-prd.md
# 2. See the graph
mcp-graph status
# 3. Get next task
mcp-graph next
# 4. Start → implement with TDD → finish
mcp-graph start
# ... write test, implement ...
mcp-graph finish
```
The AI agent does this automatically via MCP tools (`start_task`, `finish_task`, `next`).
## What's inside
| Layer | What it does |
|---|---|
| 54 MCP tools | Graph, tasks, memory, RAG, sprint, deploy |
| 130+ REST endpoints | Web dashboard, IDE integration |
| Local RAG (FTS5 + BM25) | Compressed context without external LLM |
| React dashboard | Kanban, visual graph, logs, insights |
| 9-phase lifecycle | ANALYZE → DESIGN → PLAN → IMPLEMENT → VALIDATE → REVIEW → HANDOFF → DEPLOY → LISTENING |
## Privacy
Zero telemetry. Zero phone-home. Runs offline after installation.
LLM, embeddings, and external integrations are **opt-in** — nothing active by default.
## Documentation
- [Quickstart](docs/getting-started/QUICKSTART.md) — 60 seconds
- [Full guide](docs/getting-started/GUIDE.md)
- [Cheatsheet](docs/getting-started/CHEATSHEET.md) — all commands
- [Architecture](docs/ARCHITECTURE.md)
- [Sample PRD](docs/examples/sample-prd.md)
## License
[AGPL-3.0-or-later](LICENSE) — open source with strong copyleft.
Proprietary commercial use: [commercial license available](COMMERCIAL.md).