# FeatureBoard

> FeatureBoard is a local, file-backed feature/bug board that an AI agent runs for you. It turns a project's `featurelist.md` / `buglist.md` files into a live board that Claude — or any MCP client — plans, prioritizes, and *works*: it pulls the next ticket, writes the code, logs the diff, and commits per ticket. No SaaS, no login, no lock-in. Your boards are plain markdown you own forever. FeatureBoard ships as a stdio MCP (Model Context Protocol) server and a Claude Cowork plugin.

Point FeatureBoard at a folder; each subfolder that contains a `featurelist.md` or `buglist.md` is a board. The agent reads and writes those files directly and can render the whole board as an interactive HTML artifact with a live analytics dashboard (velocity, timeline, bug health, work-log feed). The default surface is ~321 tools spanning the board and churn loop, analytics, testing, a media studio, CRM, leads, contracts, mail & marketing, a website builder with git deploys, and per-ticket commits. `FEATUREBOARD_CORE_ONLY=1` trims to the 140 core board/task tools (recommended outside Claude); `FEATUREBOARD_CLIENT_NEUTRAL=1` serves IDE-neutral instructions for non-Claude hosts.

## Install

- [Claude Cowork / Desktop plugin (v0.16.12)](https://featureboard.ai/downloads/featureboard.plugin): download and open it in Claude Cowork, then press Save to activate the MCP server plus the featureboarding skills.
- [MCP server zip for other IDEs](https://featureboard.ai/downloads/featureboard-mcp.zip): unzip and point your MCP config at `server/index.js` (requires Node.js >= 18, dependencies `@modelcontextprotocol/sdk` + `zod`). For Cursor, Grok Build, and other non-Claude hosts, set `FEATUREBOARD_CORE_ONLY=1` and `FEATUREBOARD_CLIENT_NEUTRAL=1`.
- The command every MCP client runs is the same: `node /absolute/path/to/featureboard-mcp/server/index.js` over stdio.

## Quickstart

- Install the plugin or server; on first run it asks for your boards folder (defaults to `~/FeatureBoard`).
- Say "open the FeatureBoard board" to render the live board as an artifact.
- Then just talk to it: "log a bug: gallery modal won't dismiss on mouse-out", "brainstorm five onboarding features and add them", "ship the next ticket", "how's velocity looking this week?"

## Use cases (what to ask your agent)

- "Run FeatureBoard on this repo" — bootstrap a board from an existing codebase, then pull and work tickets one at a time.
- "Break FBF-300 into subtasks and work the first one" — decompose a feature, implement it, and commit against the ticket.
- "Plan today under a token budget" — daily_plan assigns each ticket a model tier and effort level and returns a parallel dispatch plan.
- "What's on my plate / how are we looking?" — render the board with Todo / In Progress / Done columns and an analytics dashboard.
- "Reconcile logged churn vs git-actual churn" — correlate work-log diffs against real commit numstat to keep estimates accurate.
- "Build and deploy a marketing site for this project" — the website builder ships pages with git deploys.

## Optional

- [FeatureBoard home](https://featureboard.ai) — overview, interactive demo, and the roadmap.
- Measured result: a paired agent trial (n=10) showed the work packet cuts diff churn ~1.76x at equal correctness, with modestly lower token spend.
- Shipped on this board: NativeCAD (https://featureboard.ai/nativecad/), a B-rep solid-modelling kernel that reads and writes STEP, 238 KB gzipped, zero runtime dependencies, every feature and bug a ticket; and a complete CFD solver, mesher and visualizer built for $100 of tokens in under 8 hours.
- Pricing: free for private non-commercial and public/nonprofit use; commercial use is US$9.99/seat/month or US$99.99/seat/year, self-serve at https://featureboard.ai/buy after a free 24-hour trial.
