1. Conduid
  2. Data
  3. StoryFlow
MCP server · Data

StoryFlow

Visual project planning tool — Kanban board, wiki, workflow canvas, architecture view, timeline & decisions log in a glassmorphic UI

34Low

Scored 6 months ago · breakdown

About StoryFlow

StoryFlow is an MCP server in the Data category: visual project planning tool — Kanban board, wiki, workflow canvas, architecture view, timeline & decisions log in a glassmorphic UI.

Install

From source
git clone https://github.com/venom444556/StoryFlow

This server has no ConduID identity, so agent calls to it are not receipted. Pin the version you install and review the source before granting it credentials.

Ask AI

✦

Ask AI about StoryFlow

Powered by Claude · Grounded in docs

I know everything about StoryFlow. Ask me about installation, configuration, usage, or troubleshooting.

Security checks

  • ✓README presentThe repository ships documentation.

5 checks have not been run yet.

README

StoryFlow

StoryFlow is a local-first project operating system for AI-assisted software delivery. It ships three things together:

  • a React + Express app for boards, docs, workflow, architecture, timeline, decisions, and oversight
  • a terminal CLI with 60+ commands for every major StoryFlow workflow
  • a packaged StoryFlow Agent care package that can be dropped into a workspace and operated by an LLM

StoryFlow is not "an app plus some prompts." The app is the tool substrate and system of record. The agent package is the operating shell that uses it.

What Ships Today

Core product surfaces

  • Dashboard with operational project summaries
  • Overview with project operations, gates, activity feed, session history, analytics, and agent health
  • Board with epics, backlog, kanban lanes, sprints, blockers, and hygiene checks
  • Wiki with Lucide vector icons, templates, core-page auditing, and required-page enforcement
  • Workflow graph with draggable nodes, persistent positions, tidy auto-layout, and connection lines behind nodes
  • Architecture graph with dependency edges, auto-layout, draggable components, and tree view with detail slide-in
  • Timeline with phases, milestones, hot wash reports, and lessons learned rollup
  • Decisions log with immutable ADR sequence numbers and status-colored cards
  • Floating steering chat for issuing directives to the agent without blocking the viewport

Agent operating surfaces

  • storyflow context boot --json for single-call project boot with hygiene, lessons, wiki audit
  • cross-entity search and resolve with fuzzy matching
  • storyflow reconcile git for automatic board reconciliation from commit history
  • session save with all reporting fields (summary, work-done, learnings, key-decisions, next-steps)
  • batch operations: batch-done, batch-update
  • phase hot wash generation, editing, finalization, deletion, and project-level lessons rollup
  • wiki audit and required core-page creation
  • workspace-scoped packaged agent install, doctor, hook install, and status
  • full event provenance: reasoning, confidence, parent-event-id, entity-id, session-id

Agent delegation model

StoryFlow enforces autonomous PM behavior from both ends:

  • Claude pushes: dispatches the storyflow-agent when it sees work that needs tracking
  • Hooks pull: trigger automatically on commits, file edits, session start/end

The overlap is intentional. Neither side assumes the other handled it. CLAUDE.md and the agent prompt both encode this rule.

Safety and truth

  • hybrid gate enforcement for AI mutations
  • low-confidence auto-gating
  • snapshots and restore
  • event provenance with causal chains (parent_event_id)
  • session history with full context
  • server-backed analytics, gates, and operational summaries
  • database durability: debounced save on every mutation, periodic auto-save every 30s, awaited flush on shutdown, launchd ExitTimeOut protection

Quick Start

git clone https://github.com/venom444556/StoryFlow.git
cd StoryFlow
npm install
npm run dev

Open http://127.0.0.1:3001.

npm run dev starts the StoryFlow server, API, and Vite-backed UI on port 3001.

CLI

The CLI lives in cli/ and exposes both storyflow and storyflow-cli as binaries. 60+ commands, every one supports --json.

Local install from this repo

npm install -g ./cli
storyflow status

Example workflow

storyflow config set-url http://127.0.0.1:3001
storyflow config set-default saucier

storyflow context boot --json
storyflow search "auth" --json
storyflow resolve issue S-42 --json
storyflow issues create --title "Fix login" --type bug --priority high --labels "backend,auth" --json
storyflow issues done S-42 --json
storyflow issues batch-done S-43 S-44 --json
storyflow issues comment S-42 --body "Fixed in commit abc123" --json
storyflow reconcile git --json
storyflow pages audit --json
storyflow phases hot-wash lessons --json
storyflow sessions save -s "Session summary" --work-done "Fixed auth" --learnings "Token refresh needed" --json

Major command groups

  • projects, issues, sprints, board, hygiene
  • pages, decisions
  • phases, milestones
  • workflow, architecture
  • events, sessions, ai-status, gates, snapshots, steer
  • context, search, resolve, reconcile
  • agent

StoryFlow Agent Care Package

StoryFlow includes a packaged agent scaffold in agent/. The app stays the system of record, and the agent package carries the operating behavior.

Bootstrap the agent package

The agent scaffold ships in the repo. After cloning, install hooks and verify:

storyflow agent install-hooks    # wire hooks into Claude Code
storyflow agent doctor --json    # verify all 8 checks pass
storyflow agent status --json    # package health snapshot

To regenerate the scaffold from scratch (e.g. in a separate project):

storyflow agent init             # create agent/ directory
storyflow agent install-hooks

Scaffold layout

agent/
  CLAUDE.md                      # agent identity, operating rules, boot/close sequences
  SKILL.md                       # CLI quick reference
  config.json                    # hook file mapping
  hooks/
    session-start.sh             # boot: set working, context boot
    session-stop.sh              # close: set idle
    pre-mutation.sh              # gate: block commit if no In Progress issue
    post-mutation.sh             # sync: mark issues Done on commit
  kb/                            # knowledge base (core + runtime)
  memory.db                      # agent persistent memory (gitignored)
  state/                         # runtime state (gitignored)

What the care package does

  • defines the agent's operating rules, delegation model, and command path
  • boots from storyflow context boot --json
  • installs Claude-compatible hooks into .claude/settings.local.json
  • persists local runtime memory in agent/memory.db
  • runs readiness diagnostics with storyflow agent doctor
  • dual-ended delegation: agent handles PM work, hooks enforce if agent misses

Hot Wash And Lessons Learned

StoryFlow treats retrospectives as first-class project artifacts.

  • per-phase hot wash reports can be generated, edited, finalized, deleted, and listed
  • phase name resolution: reference phases by name, not just UUID
  • project-level lessons learned roll up from hot wash evidence
  • a dedicated Lessons Learned page is available in the Insights area
  • the CLI exposes storyflow phases hot-wash lessons --json
  • lessons rollup is part of the agent operating loop, not a manual afterthought

Wiki Accountability

Wiki discipline is enforced with product primitives instead of prompt reminders.

  • storyflow pages audit --json detects stale or missing core documentation
  • storyflow pages ensure-core creates missing required backbone pages
  • required core pages are shared via shared/wikiCorePages.js
  • context boot surfaces wiki freshness and missing core pages to the agent
  • all page icons use Lucide vector icons, not emojis

Architecture Overview

App and API

  • frontend: React 18, Vite 6, Tailwind CSS 4, Zustand
  • backend: Express, SQLite via sql.js, WebSocket updates
  • database durability: debounced save on every mutation, periodic auto-save every 30s, awaited flush on shutdown
  • server is the system of record for project truth
  • position-only updates exempt from rate limiting for smooth canvas drag

Agent and plugin

  • packaged local agent scaffold in agent/
  • Claude integration assets in plugin/
  • dual-ended delegation model: Claude pushes, hooks pull
  • workspace-scoped agent API:
    • GET /api/agent/status
    • GET /api/agent/doctor
    • POST /api/agent/init
    • POST /api/agent/install-hooks

Project truth endpoints

  • GET /api/projects/:id/context (agent boot contract)
  • GET /api/projects/:id/operational (UI operational summary)
  • GET /api/projects/:id/analytics
  • GET /api/projects/:id/gates
  • GET /api/projects/:id/wiki-audit
  • GET /api/projects/:id/lessons-learned
  • POST/GET/PUT/DELETE/finalize /api/projects/:id/phases/:phaseId/hot-wash

Canvas features

  • Workflow and architecture graphs with draggable nodes
  • Node positions persist to SQLite (x/y columns on workflow_nodes and architecture_components)
  • Tidy Layout button auto-arranges nodes (Sugiyama-style for architecture, BFS-layered for workflow)
  • Connection lines render behind nodes (opaque --color-bg-node backgrounds)
  • Infinite canvas — no drag boundaries
  • Auto-spread when all nodes start at (0,0)
  • Debounced position persistence with rate limit exemption

Repository Map

agent/      Packaged StoryFlow Agent scaffold
cli/        StoryFlow CLI package (60+ commands)
plugin/     Claude plugin assets, skills, and hooks
server/     Express API and SQLite data layer
shared/     Shared product contracts
src/        React app
docs/       Product docs and screenshots

Scripts

npm run dev          # Express + Vite on localhost:3001
npm run build        # Production build
npm run test         # Tests in watch mode
npm run test:run     # Tests single run
npm run lint         # ESLint
npm run ci           # Full pipeline: format + lint + typecheck + test + build

Local-First Security Model

StoryFlow is designed for local or controlled-network operation.

  • loopback-first server defaults
  • optional API token support
  • gate enforcement for AI mutations
  • snapshots before destructive changes
  • inspectable event and session history

See SECURITY.md for the security policy.

License

Copyright (c) 2026 Sheldon Spence. See LICENSE.

README copy captured when this server was indexed; it may lag the repository. The original is authoritative.

Questions

About StoryFlow

How do I install StoryFlow?

No package has been published for this server, so install it from source: clone the repository and follow its README.

Is StoryFlow safe to use with an AI agent?

Its trust score is 34 out of 100 (low). It passes 1 of the 1 static security checks run so far. It has no ConduID identity yet, so agent calls to it are not receipted.

Is StoryFlow still maintained?

Conduid hasn't recorded a commit date for this repository yet. Check the repository directly for recent activity.