Skip to content
kcjonsonPublic

About

An AI workflow based documentation and tracking platform

Resources

Contributing

Stars

0 stars

Watchers

0 watching

Forks

Latest commit

 

History

834 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Specboard

Specboard is an integrated documentation and planning platform. It combines a Git-backed markdown editor with a lightweight task manager, designed for teams that want their specs and tasks to live together.

Products

Documentation Editor

A collaborative markdown editor built for technical documentation:

  • WYSIWYG and raw markdown modes — switch between rich editing and direct markdown
  • Inline comments — Google Docs-style margin comments with reply threads
  • AI assistance — sidebar chat powered by Anthropic Claude and Google Gemini for writing help
  • Git-backed storage — documents are markdown files in a GitHub repository, with full version history
  • Real-time sync — edit in the browser, commit back to GitHub with conflict detection

Planning Board

A kanban-style task manager with document integration:

  • Epic/task hierarchy — organize work into epics with nested tasks
  • Three-column board — drag-and-drop between status columns
  • Document linking — connect specs to the epics and tasks they describe
  • MCP integration — AI coding assistants (Claude Code) can read and update tasks via the Model Context Protocol

How It Works

Specboard runs as a web application backed by AWS services. Users sign in, connect their GitHub repositories, and get a synced workspace where they can edit documents and manage tasks. Changes are stored in the cloud and can be committed back to GitHub.

For AI-assisted development workflows, the MCP server exposes planning data to tools like Claude Code, allowing AI agents to check current work, update task status, and read specs — creating a feedback loop between documentation, planning, and implementation.

Claude Code Integration

Specboard ships a plugin for Claude Code that connects your AI coding sessions to your planning board. It bundles two skills — /specboard:whats-next (discover work, scope it, break it down, keep board status accurate) and /specboard:complete (verify, finalize the PR, and close out the epic) — together with the MCP server connection, so one install wires up everything.

Setup

The user-facing version of this guide is published at specboard.io/setup.

Install the plugin:

/plugin marketplace add https://specboard.io/claude
/plugin install specboard@specboard

This registers the Specboard MCP server (https://specboard.io/mcp) and installs the workflow skill. On first connect, the OAuth flow handles authentication automatically.

In a cloud VM (Claude Code on the web, CI sandboxes): these environments need two one-time settings, because their container is rebuilt for every session and starts behind a restrictive egress proxy.

  1. Allow specboard.io in the environment's network policy. The default policy denies it, so the catalog URL above cannot be fetched and the plugin cannot be installed at all.

  2. Add a startup script so the install survives the container being reclaimed. Plugins live in ~/.claude/plugins/, which is not persisted, so without this the install is per-session:

    claude plugin marketplace add https://specboard.io/claude \
      || echo "specboard marketplace add failed (is specboard.io allowed in the network policy?)" >&2
    claude plugin install specboard@specboard --scope user \
      || echo "specboard plugin install failed" >&2

    Both commands are idempotent. --scope user installs for every repo in the VM and avoids the workspace-trust gate that applies to project-scoped plugin settings. The next session starts with /specboard:whats-next and /specboard:complete available.

Neither step involves this repository (the catalog is served from specboard.io and the plugin payload from the public npm package), so the install path is unaffected by repository visibility.

Authentication in a cloud VM: a headless session cannot complete the MCP OAuth browser redirect, so https://specboard.io/mcp will sit unauthenticated no matter what the network policy allows. Add Specboard as a claude.ai connector instead, which authenticates once at the account level and routes through claude.ai rather than the container's egress. Do not put a token in the startup script. Note that a connector does not send the X-Specboard-Project header, so a repo bound via .mcp.json will not auto-select its project; list_projects returns every project and the skill asks which to use.

Bind a repo to a project (optional): to make the skill always target one project in a given repo, commit a project-scoped .mcp.json at the repo root carrying that project's address, owner/project, the same pair you see in its Specboard URL (/projects/<owner>/<project>/planning):

{ "mcpServers": { "specboard": { "type": "http", "url": "https://specboard.io/mcp", "headers": { "X-Specboard-Project": "<owner>/<project>" } } } }

A bare <project> also works but means each caller's own project of that slug, so anyone else who clones the repo gets "project not found". Use the full form in any repo with collaborators.

This project-scoped entry overrides the plugin's server in that repo, so reconnect the MCP after adding it. On first connect it runs its own one-time OAuth — the plugin server's token does not carry over to the project-scoped one, so expect a sign-in prompt (and a trust prompt for the new server). The address is a shared reference, not a credential — each user still authenticates individually, and access is checked per user against that project.

Usage

From any project directory in Claude Code:

/specboard:whats-next

The skill also activates automatically when you're doing Specboard work. It will:

  1. Query your Specboard projects via MCP
  2. Check local git state (branches, worktrees, PRs)
  3. Cross-reference to identify active, paused, and available work
  4. Recommend what to work on next

During a session, Claude uses the Specboard MCP tools to track progress — scoping work, breaking it into tasks, keeping status accurate, linking branches, and opening (and, when verified, merging) PRs.

MCP Tools

Tool Description
list_projects Discover your projects and their IDs
get_items Query items with filtering by status, type, search; optional child and activity-log includes
create_item Create an epic, bug, or task
create_items Bulk create tasks under a work item
update_item Update status, sub-status, branch, PR URL; note appends to the item's activity log
delete_item Delete a work item or task

Documentation

Document Description
Architecture System design, infrastructure, and monorepo structure
Setup Guide Local development environment setup
Tech Stack Technology choices, coding standards, and conventions
Deployment How to deploy to staging and production

Detailed feature specifications live in docs/specs/.

Quick Start

Specboard runs entirely in Docker. No local Node.js installation required.

# Clone and start
git clone https://github.com/kcjonson/specboard.git
cd specboard
docker compose up

# Open in browser
open http://localhost

See the Setup Guide for full instructions including local overrides and database access.

License

Private — not currently open source.

About

An AI workflow based documentation and tracking platform

Resources

Contributing

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Used by

Contributors

Languages